This is the accepted manuscript made available via CHORUS.

The article has been

published as:

Symmetry-protected topological entanglement

Iman Marvian
Phys. Rev. B 95, 045111 — Published 9 January 2017
DOI: 10.1103/PhysRevB.95.045111
 Symmetry-Protected Topological Entanglement

Iman Marvian1, 2
1
Research Laboratory of Electronics, Massachusetts Institute of Technology, Cambridge, MA 02139
2
Department of Physics and Astronomy, Center for Quantum Information Science and Technology,
University of Southern California, Los Angeles, CA 90089
We propose an order parameter for the Symmetry-Protected Topological (SPT) phases which are
protected by Abelian on-site symmetries. This order parameter, called the SPT-entanglement, is
defined as the entanglement between A and B, two distant regions of the system, given that the
total charge (associated with the symmetry) in a third region C is measured and known, where
C is a connected region surrounded by A, B and the boundaries of the system. In the case of
1-dimensional systems we prove that in the limit where A and B are large and far from each other
compared to the correlation length, the SPT-entanglement remains constant throughout a SPT
phase, and furthermore, it is zero for the trivial phase while it is nonzero for all the non-trivial
phases. Moreover, we show that the SPT-entanglement is invariant under the low-depth quantum
circuits which respect the symmetry, and hence it remains constant throughout a SPT phase in
the higher dimensions as well. Also, we show that there is an intriguing connection between SPT-
entanglement and the Fourier transform of the string order parameters, which are the traditional
tool for detecting SPT phases. This leads to a new algorithm for extracting the relevant information
about the SPT phase of the system from the string order parameters. Finally, we discuss implications
of our results in the context of measurement-based quantum computation.

I. INTRODUCTION (MPS) framework, and show that it always successfully

Symmetry-Protected Topological (SPT) order is a new
kind of order at zero temperature which cannot be de-
scribed by the traditional Landau paradigm of symmetry
breaking1–4 . Different SPT orders remain distinct from
each other in the presence of an appropriate symmetry.
However, there is no local order parameter to distinguish
SPT ordered phases from the trivial phase. Also, SPT
ordered phases do not exhibit long-range order between
distant sites. First known example of a SPT phase is
the Haldane phase of the antiferromagnetic spin chains
with odd integer spins5,6 , which has been recently real-
ized in several experiments (See e.g.7–10 ). This phase is
protected by the D2 ' Z2 × Z2 symmetry corresponding
to π rotations around a set of orthogonal axis. Recently
SPT order has attracted a considerable amount of atten-
tion in quantum information community (See e.g.11–16 ).
It is often claimed that (symmetry-protected) topo-
logical order of the system is fundamentally related to
its entanglement structure. Hence, it is desirable to have
order parameters which directly detect this entanglement
structure. In fact, in the case of topological order, topo-
logical entanglement entropy can reveal the presence of
topological order based on the entanglement properties
of the ground state17–19 .
In this paper, we propose a new quantity, called SPT-
Entanglement, which detects SPT ordered phases based
on their entanglement properties. We show that for
Abelian symmetries and in the appropriate limits, SPT-
Entanglement is a universal quantity in all dimensions,
that is it remains constant throughout a SPT phase, sim-
ilar to the fact that topological entanglement entropy re-
mains constant throughout a topological phase18,19 . Fur-
thermore, in the case of 1-dimensional systems we cal-
culate SPT-Entanglement in the Matrix Product State
detects the presence of SPT order. We also show that
there is an intriguing connection between the concept
of SPT-Entanglement and the string order parameters.
More precisely, we show that SPT-Entanglement can be
expressed in terms of the Fourier transform of the string
order parameters.
Although the main results of this paper are focused on
the field of SPT order, it also contributes by showing that
the resource theory point of view to entanglement can be
useful in the study of many-body systems (See e.g.21,22
for recent reviews). In this point of view, rather than
quantifying entanglement using a particular measure of
entanglement, entanglement of a given state is charac-
terized in terms of the equivalence class of all states that
can be reversibly reached from this state via Local Oper-
ations and Classical Communication (LOCC). This point
of view to entanglement is essential for understating and
proving the properties of SPT-Entanglement. This sug-
gests that the resource theory point of view to entangle-
ment might be useful in other many-body problems.
This point of view also enables us to clearly see the ad-
vantage of using entanglement measures instead of cor-
relation measures such as mutual information. Although
in the recent years entanglement measures have been ex-
tensively used in the study of many-body systems, it is
not always clear that why they are relevant in this con-
text, and why they cannot be replaced by other mea-
sures of correlation, such as mutual information, which
are often easier to calculate. Interestingly, in the argu-
ments presented in this paper, one can clearly see that the
monotonicity of measures of entanglement under classical
communication, which distinguishes them from general
measures of correlations, is crucial for having a univer-
sal quantity which remains constant throughout a SPT
phase.
 2

A. Outline G. Note that the charges are defined only up to a 2π

shift.The number of distinct irreps of an Abelian group
We start by reviewing some preliminary concepts. In G are equal to |G|, the order of group. Therefore, the
particular, in Sec.(II A) we review the notions of symme- set of distinct charges, denoted by Q, has also |G| differ-
try and charge, and we discuss the example of Z2 × Z2 ent elements. Note that this set itself forms an additive
symmetry. Then, in Sec.(II B) we briefly review the no- group. That is for any charges κ1 , κ2 ∈ Q, we have
tion of low-depth symmetric circuits. Next, in Sec.(II C)
we review some basic concepts of entanglement theory, κ1 + κ2 = κ3 (mod 2π) , (2.2)
namely the concepts of measures of entanglement and for some κ3 ∈ Q.
LOCC protocols. Then, in Sec.(III) we define the no- Recall the orthogonality relations between 1-
tion of SPT-entanglement, and present a summary of dimensional irreps of Abelian groups,
our main results. In Sec.(III C) we show that there is
X
a connection between the notion of SPT-Entanglement eiκ1 (g) e−iκ2 (g) = |G| × δκ1 ,κ2 , (2.3)
and the string order parameters. Next, in Sec.(IV) we g∈G
consider the example of perturbed cluster Hamiltonian,
and calculate the SPT-Entanglement for its ground state, where δ is the Kronecher delta function. Using these
which is in the SPT phase protected by Z2 ×Z2 symmetry. relations, we find that the projector to the subspace with
Then, in Sec.(V) we focus on the case of 1-dimensional charge κ can be written as
systems, and calculate the SPT-Entanglement using the
MPS classification of SPT phases. Finally, in Sec.(VI) we 1 X −iκ(g) (C)
Π(C)
κ = e U (g) . (2.4)
present the formal versions of our general results on SPT- |G|
g∈G
Entanglement, based on the notion of low-depth symmet-
ric circuits. It can be easily shown that the total charge in a region
can be written as the sum of the charges in the subregions
of this region. In particular, if we partition C into two
II. PRELIMINARIES non-overlapping subregions C1 and C2 such that C =
C1 ∪ C2 , then
A. From symmetry to charge X
Π(C)
κ = Πκ(C1 1 ) ⊗ Π(C
κ2
2)
, (2.5)
Consider a finite-dimensional lattice system. Let G be κ1 ,κ2 ∈Q
κ1 +κ2 =κ (mod 2π)
an Abelian symmetry which has a unitary on-site rep-
resentation on this lattice. In particular, let ul (g) be where the summation is over all charges κ1 , κ2 ∈ Q which
the (linear) unitary representation of the group element add up to κ (mode 2π). Note that the charges corre-
g ∈ G on site l, such that ul (g1 )ul (g2 ) = ul (g2 )ul (g1 ) = sponding to non-Abelian groups are not additive.
ul (g1 g2 ) for all g1 , g2 ∈ G. Note that the symmetry G Let
can be an Abelian subgroup of a non-Abelian symmetry O X
of the Hamiltonian. U (g) = ul (g) = eiκ(g) Πκ (2.6)
Let C be an arbitrary subset of sites on this lattice. l∈C κ∈Q
Then, the action of the group element g N ∈ G on C
is represented by the unitary U (C) (g) = be the representation of the group element g ∈ G, on
l∈C ul (g).
Since all unitary Irreducible Representations (irreps) of all sites in the system, and Πκ be the projector to the
an Abelian group are 1-dimensional, this representation subspace with the total charge κ in the system. We call
can be decomposed as a unitary transformation V symmetric if it commutes
O X with the action of the symmetry group, that is
U (C) (g) = ul (g) = eiκ(g) Π(C)
κ , (2.1)
l∈C κ ∀g ∈ G : U (g)V U † (g) = V , (2.7)

where each eiκ(g) is a 1-dimensional representation (char-
(C)
acter) of the group G, and Πκ is the projector to the
corresponding subspace.
As we will see in the following sections, instead of
thinking in terms of 1-dimensional irreps eiκ(g) , it is
sometimes more useful to consider the functions κ(g) cor-
responding to the arguments of the 1-dimensional irreps,
and interpret them as different charges. Therefore, in
(C)
the above decomposition each projector Πκ can be in-
terpreted as the projector to the sector with charge κ in
region C. Each charge κ is a real function over the group
or equivalently, [V, U (g)] = 0. This definition implies
∀κ ∈ Q : V Πκ V † = Πκ , (2.8)
which basically means that the total charge in the system
is conserved under symmetric unitaries.
As we will see in the rest of the paper, this conserva-
tion law, together with the additivity relation in Eq.(2.5),
which holds only for Abelian groups, play important roles
in our arguments. Indeed, a useful advantage of think-
ing in terms of additive charges instead of 1-dimensional
irreps, is that one can use the standard intuition about
 We come back to this example in Sec.(IV), where we
discuss SPT
Di↵erent SPTphase phases of cluster
can bestate. classified based on the
equivalence classes of states induced by the symmetric 3
24
low-depth quantum circuits . According to this classifi-
ro
the additivity and the conservation of electrical charges Note that I+(−1)2 X odd
is the projector to the subspace
cation,for two B.
states Symmetric
are in theofsame low-depth
SPT phase circuits
to find a better understanding of the SPT phases. Note which the total parity X measurements onifallone
oddcan be
re
that even if the on-site symmetry of the Hamiltonian approximately
is transformed
qubits in region to the other2 by a islow-depth
C is ro . Similarly, I+(−1) X
even
the
non-Abelian, we can choose G to be an Abelian circuit sub- Di↵erent
projector
Qlto the subspace for which the total parity of
V = k=1 SPT phaseseach
Vi where can Vbe i isclassified
a productbased ofXa set
on ofthe
group of this symmetry, and apply the arguments for the measurements on all even qubits in region C is re .
charges corresponding to this Abelian subgroup. unitaries which classes
equivalence (i) act locally
of states
We come back to this example
on induced
non-overlapping regions
by the symmetric
in Sec.(IV), where we
of the system, and (ii) are 24
invariant under the symme-
low-depth quantum circuits
discuss SPT phase of cluster state. . According to this classifi-
try cation,
(That two is satisfy
states [V are i, Uin(g)]
the same= 0, SPT for allphaseg 2ifG). Thebe
one can
1. Example: Z2 × Z2 symmetry
circuit should be low-depth
approximately transformed in the to the sense thatbythe
other depth l
a low-depth
Ql Symmetric low-depth circuits
B.
times
In this section we consider the example of the groupcircuit V =
the maximum range of each unitary in the circuit is
k=1 Vi where each Vi is a product of a set of
bounded
unitaries by some
which range actRlocallywhichon in non-overlapping
the thermodynamics
Z2 ×Z2 . Formally, this group can be thought as the group Different SPT(i) phases can be classified based on the regions
of strings of two bits bo be with bitwise XOR as the group limit is
of the negligible
system,
equivalence compared
andof(ii)
classes states to the
areinduced
invariant system
by the undersize.the symme-
symmetric
operation, that is the set {00, 01, 10, 11}. This groups An important feature of the low-depth circuits
g 2 is G).that
20
low-depth
(That isquantum satisfycircuits . According to thisallclassifi-
protects the Haldane phase of spin chain with spin-onetry cation, two states are
[Vi , U (g)]
in the same SPT
= 0, for
phase if one can be
The
6they have a light cone. That is if the circuit V has range
systems, which includes states such the AKLT state . Incircuit should be
approximately low-depth
transformed to in
the the othersense that the depth l
by a low-depth
this case the on-site representation of this group isR, thethen
times for any
theVmaximumlocal
Ql operator
rangeeach O
of Veach
a whichunitary support
has only is
in the circuit
circuit = Vi where i †is a product of a set of
set of four operators {I, eiS π , eiS π , eiS π }, where I isonthea single site a,k=1
unitaries whichthe operator
locally onVnon-overlapping
Oa V has support regions only
x y z

identity operator, and Sx , Sy and Sz are spin matrices inbounded by some range R which in the thermodynamics
(i) act
three orthogonal directions. Note that for integer spinslimit is negligible compared to the system R.
on sites whose distance from site a is, at most, In other
of the system, and (ii) are invariant under the symme-
size.
words,An try
under(That is satisfy [V , U (g)]
this type of transformations
i = 0, for all g ∈ G). The
local informa-
the π rotations around three orthogonal axes commute important
circuit should be feature
low-depthof in the
the senselow-depth
that the circuits
depth l is that
with each other. tiontheycannothave
times propagate
theamaximum
light cone. outside
rangeThat a region
of eachisunitary with
if theincircuit diameter
the circuit of
V ishas range
As another example, consider the representation oforder this R (See Fig.(1)) .
bounded by some range R, which in the thermodynamics
group on a spin chain with spin-half systems (qubits). WeR, then for any local operator Oa which has support only
limit is negligible compared to the system size.
can group pairs of neighbor qubits, and consider each pairon a single site a, the operator V † Oa V has support only
as one site. Then, on each site formed from two qubits, An important feature of the low-depth circuits is that
the group Z2 × Z2 can be represented by
on sites
they whose
have distance
a light cone. That from is ifsite
theacircuit
is, atVmost, R. In other
has range
C.R,
Measures
then for any
ofoperator
local
Entanglement O which has
and
support
LOCC only
words, under this type of transformations a local informa-
u(bo be ) = X b ⊗ X b , be,o ∈ {0, 1} , a single site a, the operator V † Oa V has support only
(2.9)tiononcannot
o e

on sites whose propagate

distance from outsidesite a ais,region
at most,withR. diameter of
where X denotes the Pauli σ operator. In other words,
order R (See Fig.(1)) .
x
on each pair of qubits the group is represented by one of A A0
the following operators
C. years
In recent Measures of Entanglement
entanglement theory inand LOCC
general-and
{I ⊗ I, X ⊗ I, I ⊗ X, X ⊗ X} .
measures of entanglement in particular- has found lots of
Let C be a connected region with even number of
qubits 2l. Then, the action of group in this region is
represented by
A A0
bo
U (C) (bo be ) = Xodd be
Xeven , (2.10)
InFIG.recent
1: The years
light-coneentanglement
corresponding to atheory
low-depthincircuit.
general-and
where Xodd = (X ⊗ I)⊗l (or Xeven = (I ⊗ X)⊗l ) are themeasures After applying the unitary, information which is initially lo-
of entanglement in particular- has found
calized in region A, will stay inside region A0 , which is only
lots of
Pauli σx operators acting on all odd (or even) qubits in
slightly larger than A.
region C.
This Abelian group has four elements, and hence has
four different unitary 1-dimensional irreps, or equiva-
lently four different charges. These four different charges
can be labeled by two bits ro , re ∈ {0, 1}. In particular,
for the charge labeled by ro re the group element bo be is C. Measures of Entanglement and LOCC
represented by (−1)ro bo +re be . Then, the projector to the
subspace with this charge in region C is In recent years entanglement theory in general-and
measures of entanglement in particular- has found lots of
1
1 X bo applications in many-body systems (See e.g.24–28 ). Many
Π(C)
ro re = (−1)bo ro +be re Xodd be
Xeven (2.11a)
4 authors have studied how different measures of entan-
bo ,be =0
glement, such as negativity23 or concurrence32 , can de-
I + (−1)ro Xodd I + (−1)re Xeven tect certain properties of a many-body system. In this
=( )( ) . (2.11b)
2 2 paper, however, we are not concerned with particular

measures of entanglement, because our results on SPT-
Entanglement hold for all such functions. Indeed, it turns
out that to understand SPT phases, and in particular,
the notion of SPT-Entanglement, it is useful to go be-
yond particular measures of entanglement, and take a
more resource-theoretic perspective to entanglement.
In quantum information theory entanglement is de-
fined as the property of states that cannot be gener-
ated with Local Operations and Classical Communica-
tions (LOCC). These are basically all and the only trans-
formations that can be implemented on a bipartite sys-
tem shared between two distant local parties, Alice and
Bob, if the only interactions between them is through a
classical channel. These transformations include (i) local
operations on each side, such as local unitary transforma-
tions, measurements, and discarding subsystems (partial
trace) and (ii) classical communication between the two
parties. Under this restricted class of operations, an un-
tangled state cannot be transformed to an entangled one.
Also, a given entangled state cannot be transformed to
another arbitrary entangled state, if the latter state has
more entanglement.
Recall that a general bipartite mixed state ρAB is un-
entangled,
P or separable, if it can be written as ρAB =
A B
i pi ρi ⊗ ρi , that is a convex combination of uncor-
related states ρA B
i ⊗ ρi , where {pi } is a probability dis-
tribution. Note that using LOCC transformations it is
possible to transform
P an uncorrelated state ρA ⊗ ρB , to
a correlated one i pi ρi ⊗ ρB
A
i , using LOCC a state with
zero mutual information can be transformed to a state
with non-zero mutual information.
Unlike mutual information which can increase under
classical cimmunication, measures of entanglement can
distinguish between classical correlations and quantum
correlations which cannot be generated via classical com-
munication. In particular, if using LOCC a bipartite
state ρAB can be transformed to state σ AB , then any
measure of entanglement E should be non-increasing in
this transformation, i.e.
LOCC
ρAB −−−−→ σ AB =⇒ E(ρAB ) ≥ E(σ AB ). (2.12)
LOCC
Here, the arrow −−−−→ means that the transformation
from the first state to the second is possible via Local
Operations and Classical Communication.
In this paper, rather than focusing on a particular
measure of entanglement, we phrase our results in terms
of interconvertability of states under LOCC. Note that
monotonicity of measures of entanglement under LOCC,
i.e. Eq.(2.12), immediately implies that
Proposition 1 Suppose a pair of states ρAB and σ AB
can be reversibly transformed to each other using LOCC,
LOCC LOCC
that is ρAB −−−−→ σ AB and σ AB −−−−→ ρAB . Then,
any measure of entanglement E takes the same value on
these two states, i.e. E(ρAB ) = E(σ AB ) .
In addition to the monotonicity under LOCC, it is usu-
ally assumed that a measure of entanglement vanishes on
4
all unentangled (separable) states. In particular, for all
product states ρA ⊗ ρB , it should hold that
E(ρA ⊗ ρB ) = 0 . (2.13)
Note that because any unentangled state can be trans-
formed to any other unentangled state via LOCC, propo-
sition 1 implies that all measures of entanglement should
take the same value on all unentangled states. Therefore,
Eq.(2.13) can be thought as just a convention that fixes
this value to be zero.
A well-known example of measures of entanglement is
negativity 23 defined by
TA
AB kρAB k1 − 1
N (ρ )≡ , (2.14)
2
T
where ρAB A is the operator obtained by partial trans-
pose on system A (relative to an arbitrary basis), and
k · k1 is the sum of the absolute values of the eigenvalues
of the operator. One can easily check that for a maxi-
mally entangled state of a pair of d-dimensional systems,
Pd
i.e. state √1d i=1 |iii the negativity is equal to (d−1)/2.
Note that the entanglement entropy of pure bipartite
states, defined as the entropy of the reduced state of one
side, is non-increasing under LOCC in pure state to pure
state transformations. However, it can increase in the
transformations between mixed states, and therefore it
is not a valid measure of entanglement for mixed states.
For instance, for the totally mixed state of a bi-partite
system, which is clearly unentangled, the entropy of the
reduced states could be arbitrarily large, depending on
the dimension of the Hilbert space.
1. Example: Conditionally rotated states
Consider a bi-partite pure state |ψiAB shared be-
tween two distant parties, Alice and Bob. For any lo-
cal unitary U acting on the Bob’ system, the two states
(I ⊗ U )|ψiAB and |ψiAB , have the same amount of en-
tanglement relative to any measure of entanglement E,
that is E((I ⊗ U )|ψiAB ) = E(|ψiAB ).
Now suppose {Uk } is an arbitrary set of unitary trans-
formations acting on system B, and {pk } is a probability
distribution over this set. Consider the convex combina-
tion
X
ΩAB = pk (I ⊗ Uk )|ψihψ|AB (I ⊗ Uk† ) , (2.15)
k
that is a mixture of rotated versions of state |ψiAB ,
weighted by the probability distribution {pk }.
Then, in general, for any measure of entanglement E
it holds that
   
E ΩAB ≤ E |ψihψ|AB . (2.16)
This is because state |ψiAB can be transformed to state
ΩAB via local operations, namely by applying unitary

Uk on system B with probability pk . In general, the left-
hand side is strictly smaller than the right-hand side. In
particular, we can always choose unitaries {Uk }k and a
probability distribution pk such that this state is unen-
tangled, for all initial states |ψiAB .
Next, consider state
X charge in C the charge κ is obtained with the proba-
ΩABK = pk |kihk|K ⊗ (I ⊗ Uk )|ψihψ|AB (I ⊗ Uk† ) .
k duced state of AB after the charge measurement will be
(2.17)
Here, states {|ki} are orthonormal states of a classical
register K, which keeps information about the unitary
Uk which is applied to system B. Now suppose we give
the systems K, and B to Bob, and system A to Al-
ice. In other words, we partition these three systems
as A|KB. Then, it can be easily seen that the bi-partite
entanglement of state ΩABK relative to this partition is
still equal to the entanglement of state |ψiAB . In other
LOCC
words, both transformations |ψiAB −−−−→ ΩABK and
LOCC
ΩABK −−−−→ |ψiAB can be implemented via LOCC. In
particular, to perform the latter transformation, Bob can
measure the value of register K, and if he finds the reg-
ister in state k, then he applies unitary Uk† on system B.
We conclude that for any measure of bi-partite entangle-
ment E for the partition A|KB, it holds that
E(ΩABK ) = E(|ψiAB ) . (2.18)
One can easily check this formula for negativity N , and
show that it takes the same value on both states.
Next, suppose instead of partition A|KB, we consider
the partition AK|B. That is we give the classical register
K to Alice, instead of Bob. What is the entanglement
of state ΩABK relative to this partition? It can be eas-
ily seen that the entanglement remains unchanged. The
reason is that register K has only classical information,
and therefore Alice and Bob can transfer this information
from one side to the other via classical communication.
Note that in the above argument about the equivalence
of bi-partite entanglement relative to A|KB and AK|B
partitions, the monotonicity of entanglement measures
under classical communication plays an important role.
If instead of measures of entanglement, we consider a gen-
eral measure of correlation, such as mutual information,
which can increase under classical communication, then
the bi-partite correlation in state ΩABK will not be gen-
erally the same relative to these two different partitions.
III. SPT-ENTANGLEMENT
A. Definition
Let A and B be two non-overlapping regions of the
system, and C be a connected region surrounded by A,
B and the boundaries of the system (see Fig 2). Let G
be an Abelian symmetry which has on-site linear unitary
representations on the sites of this system. Suppose we
5
measure the total charge associated with the symmetry
G in region C and obtain charge κ. Here, each κ cor-
responds to a distinct irreducible representation of G as
(C)
g → eiκ(g) . Let Πκ be the projector to the subspace
of states with charge κ in region C (See Eq.(2.4)). If
the system is in state ρ, then by measuring the total
(C)
bility pκ = tr(ρΠκ ), and given this outcome, the re-
(AB) (C)
ρκ ≡ trAB (Πκ ρ)/pκ , where the trace is over all the
sites in the system except those in A and B.
C
A C B B
A
FIG. 2: Region C is surrounded
KCby regions A and B and the
boundaries of the system.
A B
Consider the state
X
Ω(AB|C) (ρ) ≡ pκ |κihκ|(KC ) ⊗ ρ(AB)
κ (3.1a)
κ
X
= |κihκ|(KC ) ⊗ trAB (Π(C)
κ ρ) , (3.1b)
κ
where {|κi} are the orthonormal states of a classical reg-
ister KC which keeps the information about the outcome
of charge measurement in region C. State Ω(AB|C) (ρ)
describes the average joint state of regions A and B and
the register KC after the charge measurement.
Note that the super-operator
X
Ω(AB|C) (·) ≡ |κihκ|(KC ) ⊗ trAB (Π(C)
κ ·) (3.2)
κ
is a trace-preserving completely positive map, i.e. a
quantum channel. This is basically the map which traces
over all the degrees of freedom in the system except (i)
the total charge in region C, and (ii) local degrees of
freedom in regions A and B. Therefore, an alternative
interpretation of state Ω(AB|C) (ρ) is that it is the reduced
state of the system relative to the algebra generated by
the local observables in regions A and B and the observ-
able corresponding to the total charge in region C.
Now we are ready to define the notion of SPT-
Entanglement:
Definition 2 Let A and B be two non-overlapping re-
gions of the system, and C be a connected region sur-
rounded by A and B and the boundaries of the system
(see Fig 2). The SPT-Entanglement of state ρ between
A and B relative to C is defined as the bipartite entan-
glement of state Ω(AB|C) (ρ) relative to AKC |B partition,
or equivalently, relative to A|KC B partition (See Fig 3).
 A
KC
A B
FIG. 3: State Ω(AB|C) (ρ) describes the joint state of the lo-
cal degrees of freedom at regions A and B, and register KC ,
which corresponds to the total charge in region C. The SPT-
Entanglement is defined as the bipartite entanglement of this
state relative to the partitions A|KC B or AKC |B.
Several important remarks are in order: First, since KC
is a classical register, it can be transferred from one local
party to the other via classical communication. There-
fore, the bipartite entanglement of state Ω(AB|C) (ρ) rel-
ative to the partitions AKC |B and A|KC B is the same.
Second, we are interested in the SPT-Entanglement in
the limit where regions A and B are large and far from
each other. Third, the Abelian group G could be a sub-
group of a (possibly non-Abelian) group that protects the
phase.
Note that in this definition we talk about bipartite
entanglement of state Ω(AB|C) (ρ), without specifying a
particular measure of entanglement. This is because
the results presented in this paper hold for all measures
of entanglement, and we do not need to specify one.
But, to have an explicit example, we can quantify SPT-
Entanglement using negativity, which yields the function
N (AB|C) (ρ) ≡ N (Ω(AB|C) (ρ))
"
h #
1 X
i TA
charges in regions ∆A and ∆B, and adding these charges
(C)
to the charge in region C 0 , whose value is given to them,
= tr (Π ρ)
2 κ
AB κ
− 1 , (3.3)
1 Alice and Bob can find the charge in region C for the
where TA denotes the partial transpose, and k · k1 is
the trace norm, i.e. the sum of the absolute value of
eigenvalues. As we will see in Sec.(III C), this function
can be expressed in terms of the string order parameters.
B. Main Properties: Summary of results
In this section we present a summary of the main prop-
erties of the SPT-Entanglement.
(i) It is clear from the definition of SPT-Entanglement
that applying a symmetric unitary which acts non-
trivially only on one of the three regions A, B and C,
does not change the SPT-Entanglement. On the other
hand, a symmetric unitary with support in more than
one region, can change the SPT-Entanglement. How-
ever, we prove that if the symmetric unitary is a low-
depth circuit then the SPT-Entanglement cannot change
drastically. More precisely, we show that the effect of
a low-depth symmetric circuit on the SPT-Entanglement
can be bounded based on the effect of small deformations
of the boundaries of regions A and B on this quantity:
If by deforming the boundaries of these regions the SPT-
Entanglement cannot change considerably, then it also
cannot change considerably under low-depth symmetric
circuits (See theorem 11). Note that SPT-Entanglement
can change drastically under low-depth non-symmetric
unitaries. In the following, we explain the intuition be-
hind this result. See Sec.(VI B) for a more formal (and
slightly different) argument.
Suppose a symmetric low-depth circuit V is applied on
a system in the initial state ρ, and transforms it to state
V ρV † . Under this transformation quantum information
and charge from region A can leak to region ∆A, which
is a subset of sites in region C which are close to the sites
in region A (See Fig.(4)). Similarly, the charge and quan-
tum information in region B can leak to ∆B, a subset
of sites in region C which are close to the sites in region
B. Let C 0 = C \ (∆A ∪ ∆B) be the set of sites in region
C which are outside ∆A and ∆B, and therefore are not
close to A or B. Then, under the effect of circuit V the
charge from region C 0 can also leak to regions ∆A and
∆B (See Fig.(4)).
Now suppose Alice is given all the sites in region
A0 = A ∪ ∆A, and Bob is given all the sites in region
B 0 = B ∪ ∆B. Furthermore, suppose they know the
charge in region C 0 (for state V ρV † ). In other words,
0 0 0
assume they are given state Ω(A B |C ) (V ρV † ). Then, by
applying a symmetric unitary on region A0 , Alice can
undo the effect of V on A and restore the leaked charge
and quantum information back to this region. Similarly,
by applying a local symmetric unitary on B 0 , Bob can
restore the charge and quantum information leaked from
region B back to this region. Next, by measuring the
original state ρ, and construct the state Ω(AB|C) (ρ). This
proves that0 there exists a LOCC protocol which trans-
0 0
forms Ω(A B |C ) (V ρV † ) to state Ω(AB|C) (ρ), which, in
turn, implies that for any measure of entanglement E,
0 0
|C 0 )
E(Ω(A B (V ρV † )) ≥ E(Ω(AB|C) (ρ)) , (3.4)
where the entanglement can be considered relative to
A|KC B partition or AKC |B partition.
Similarly, using the fact that if V is a symmetric
low-depth circuit, then V † is also a symmetric low-
depth circuit, we can find an upper bound on the SPT-
Entanglement of state V ρV † for regions A0 and B 0 , based
on the SPT-Entanglement of the original state ρ for re-
gions which are slightly larger than A0 and B 0 .
(ii) According to the classification of SPT phases based
on the symmetric low-depth circuits20 , the trivial phase
is the set ofN
states which can be transformed to a product
state ρ = i ρi using such circuits. As we saw above,
the effect of low-depth symmetric circuits on the SPT-
Entanglement can be bounded by the effect of small de-
formations of the boundaries of regions A and N B on this
quantity. But, clearly, for a product state i ρi the
SPT-entanglement between any non-overlapping regions
eer chosen
these to be symmetric. Therefore, the total charge
in region A and the total charge in region B remain con-
region
ved and A servedthe
and undertotal
these charge
unitaries.in region
Since (UB remain
UB )V has con- only
A
rved
he SPT under theseon
support unitaries.
C̃ and since Since (UA Uwith
it commutes B )Vthe has only
symmetry
mmetric on it
upport C̃ follows
and sincethat ititcommutes
commutes withwith the symmetry
the total charge opera- 7
sform
followsit that
tor in
it the region C̃. with
commutes Thesethe two total
facts about
charge (UAopera- UB )V
ect SPT imply that
orregions
in the region C̃. given
These wetwomeasure
factstheabout total charge
(UA in U region
B )V
C̃
and obtain a givenA charge the post-measurement C joint that SPT-Entanglement should remain invariant under
mply
e and so that given we measure the total charge
state of Acore and Bcore will be the same for two states
in region C̃ symmetric low-depth circuits, and therefore is constant
ndrentobtain
re- a given
|⇤⌃ and (U charge
A UB )Vthe|⇤⌃ post-measurement
= (UA UB )|⇤ ⌃, i.e. joint for all throughout a phase.
ate of Acore
glement and Bcore will be the same for two states
charges Note that in the case of 1-dimensional systems explicit
s onand
⇤⌃ the (U UB )V |⇤⌃1 = (UA UB )|⇤ ⌃,⇥ i.e. for all
A calculation of SPT-Entanglement in the MPS framework
f size of ⇥(Acore ,Bcore ) ⇥ tr (C̃)
|⇤⌃⇧⇤| (C̃) (0.3)
harges C’
p̃ C̃ A confirms this assumption. However, the validity of this
1 ⇥ ⇥
ume the 1 ( C̃) † † ( C̃) assumption in the higher dimensions is not clear, and
Acore ,Bcore )
general ⇥ trC̃ = p̃(C̃) |⇤⌃⇧⇤| [U(AC̃) UB ]|⇤ ⌃⇧⇤ |[UA(0.3)
trC̃ UB ]
remains an open problem.
e of the
p̃
FIG. 4: Under the symmetric low-depth circuit V quantum
1 p̃ ⇥ ⇧⇤| (C̃) (C̃) ⇥ (iv) We directly calculate SPT-Entanglement of 1-
undaries whereinformation and
(C̃) charge
|⇤⌃ = ⇧⇤ |[UA† region
from UB†
†] A can [U†Aleak to region ∆A,
= trC̃ [UA UB ]|⇤ ⌃⇧⇤ |[UA UB ] (C̃) dimensional systems in the Matrix Product State (MPS)
ily show that
UB ]|⇤ is a we
p̃ ⌃. Now subset
showof thatsites thisinequation
region implies
C which thatare close to the
of A and the sites
SPT entanglement
in region A. between Applying A and
the B relative tounitary
symmetric C UA on the framework, and show that in the limit where A, B and
he SPT is the same 0 two states |⇤⌃ †and |⇤ ⌃.
for Here,
(C̃) we again
here p̃ ⇥ region
⇧⇤|
(C̃)
A |⇤⌃= A=∪ ⇧⇤ ∆A,|[U Alice can U † restore
] [U the
A
leaked charge and C are large compared to the correlation length, SPT-
A B
]|⇤ ⌃. Now quantum
we show information
that this from region
equation A back
implies that to this region. Entanglement is: (1) independent of the sizes of these
B
he SPT entanglement between A and B relative to C regions, (2) constant throughout all SPT phases, and (3)
the same for two states |⇤⌃ and |⇤ ⌃. Here, we again zero if and only if the system is in the trivial phase. Fur-
thermore, we show that the value of SPT-Entanglement
A0 and B 0 is zero. It follows that, in the limit where two in the SPT ordered phases is related to a physical prop-
regions A and B are far from each other compared to the erty of the SPT phase, namely the edge mode degeneracy
correlation length, the SPT-Entanglement is always zero associated with each edge of an open chain. More pre-
in the trivial phase. Applying this general result to the cisely, we show that in the limit where A and B are large
negativity, as an example of measures of entanglement, and far from each other the SPT-Entanglement is equal
we find to the entanglement of a maximally entangled state of a

i TA

X
h
pair of d[ω] -dimensional systems, where d[ω] is the dimen-
Trivial phase =⇒ lim
tr (Π(C) ρ)
−1 = 0 ,

AB κ
sion of the projective irreducible representations of the
κ 1
group G in the cohomology class [ω] which characterizes
(3.5)
the SPT phase of the system1–4 . Note that, in general,
where TA denotes the partial transpose, and k · k1 is the
irreducible representations which belong to the same co-
trace norm, and we are looking to the limit where A and
homology class have different dimensions. However, in-
B are far from each other.
terestingly, we find that for Abelian groups all such ir-
Therefore, a non-vanishing SPT-Entanglement be-
reducible representations have the same dimensions (See
tween two distant regions, in the limit where the distance
lemma 5). It turns outs that the parameter d[ω] has a
between them is large, indicates the presence of SPT or-
natural interpretation as the edge mode degeneracy as-
der. We show that this method can always detect all non-
sociated with each edge of an open chain11,29 (Note that
trivially SPT ordered phases in 1-dimensional spin sys-
this parameter is only defined for 1-dimensional systems).
tems. Another way to interpret this result is that it gives
a criterion that should be satisfied by any state which can According to the classifications of the SPT phases of
be prepared by symmetric low-depth circuits. Note that, 1-dimensional systems, the equivalence class [ω] uniquely
this method works even in the absence of translational determines the SPT phase1–4 . Therefore, in the case
symmetry (See40 for a different approach). of 1-dimensional systems direct calculation of SPT-
(iii) The above result implies that the SPT- Entanglement confirms our general result that SPT-
Entanglement remains constant throughout the trivial Entanglement is constant throughout a SPT phase.
phase. Making an assumption about the decay of cor- It follows from this result that, using different measure
relations in the system, we can extend this result to all of entanglement to quantify SPT-Entanglement we can
SPT phases, and show that in the limit where regions construct different order parameters to detect SPT or-
A and B are large and far from each other compared der of 1-dimensional systems. In particular, suppose we
to the correlation length, SPT-Entanglement is a uni- quantify the SPT-Entanglement using negativity. Then,
versal quantity in all dimensions, i.e. it remains con- since the negativity of a maximally entangled state of a
stant throughout the phase (See corollary 13). Similar pair of d[ω] -dimensional systems is (d[ω] − 1)/2, this re-
to the argument that shows topological entanglement en- sult implies that in the limit where A, B and C are much
tropy remains constant throughout a phase18,19 , our ar- larger than the correlation length, lim N (AB|C) (ρ) =
gument is based on a reasonable, but unproven assump- (d[ω] − 1)/2, or equivalently
tion: Since SPT-Entanglement increases monotonically X


h i TA



with the sizes of regions A and B (See proposition 10), lim
tr (Π(C) ρ)
− 1 = d[ω] − 1. (3.6)

AB κ

given that the correlations are short range and the system κ 1
is sufficiently homogenous (e.g., translational invariant),
it seems reasonable to assume that in the limit where re- A 1-dimensional system is in a non-trivial phase iff the
gions A and B are large, SPT-Entanglement is saturated, corresponding cohomology class [ω] is non-trivial1–4 , or
i.e. it does not increase if we make A and B a bit larger. equivalently iff d[ω] > 1. Therefore, the left-hand side
Then, this assumption together with our result (i) imply of Eq.(3.6) can serve as an order parameter whose value
 8

for the large blocks of A, B, and C not only detects the string order parameters. In particular, if we use the neg-
presence of SPT order, but also reveals the dimension ativity of entanglement, we find
of the irreducible representations in the equivalence class
[ω] which characterizes the phase. For example, in the 1 X

X ∗ (A)


(B) T

N (AB|C) (ρ) = [
s̃k,l (κ) [Xk ⊗ Yl ]
− 1] ,
case of Haldane phase where d[ω] = 2, the left-hand side 2 κ 1
k,l
of Eq.(3.6) converges to 1, while it converges to zero in (3.11)
the trivial phase. In Sec.(IV) we check this result in the where T is the transpose relative to an arbitrary basis. As
case of the cluster Hamiltonian. we explained before, in the case of 1-dimensional systems,
and in the limit where A, B and C are large this quantity
converges to (d[ω] − 1)/2, where d[ω] is the dimension of
C. Connection with the string order parameters
the irreps in the cohomology class [ω] that characterizes
the SPT phase.
Remarkably, it turns out that the notion of SPT- We conclude that, although each single string order pa-
Entanglement is closely related to the concept of string rameter skl (g) alone does not have any information about
order parameters, which have been the traditional tool the SPT phase of the system33 , but together they provide
for detecting the SPT order9,10,31,33–36 . enough information to find the SPT-Entanglement, and
The string order parameters are the expectation values hence detect the SPT order. Indeed, even further, they
of state for string operators, in the following form can be used to find the dimension d[ω] corresponding to
(A)
O (B) the phase of system.
skl (g) ≡ Tr(ρ Xk uj (g) ⊗ Yl ) , (3.7a) See31 for another recent approach to string order pa-
j∈C rameters, which applies for Abelian groups, provided that
(A) (B) the cohomology class describing the phase satisfies a con-
= Tr(ρ Xk ⊗ U (C) (g) ⊗ Yl ) (3.7b)
dition called maximal non-commutativity.
(A) (B)
where {Xk }k and {Yl }l are basis for the space of
local operators acting on A and B respectively, and g →
uj (g) is the representation of the group element g ∈ G on D. Remarks on the definition of SPT-Entanglement
site j. The relation between the string order parameters
and the SPT-Entanglement can be established using the Can we modify the definition of SPT-Entanglement
notion of Fourier transform for Abelian groups. Let and keep its nice properties? There are several assump-
tions in the definition of SPT-Entanglement, and one
1 X −iκ(g) may wonder if these assumptions are crucial or not. Here,
s̃kl (κ) = e skl (g) , (3.8)
|G| we discuss more about the importance of these assump-
g∈G
tions.
be the Fourier transform of the string order parameter
(C) P
skl (g). Then, using Πκ = |G|−1 g∈G e−iκ(g) U (C) (g),
and the definition of state Ω(AB|C) (ρ) in Eq.(3.1), we can 1. Why Abelian symmetries?
easily see
h i  In both arguments that we present in the paper, i.e.
(A) (B)
s̃kl (κ) = Tr |κihκ|(KC ) ⊗ Xk ⊗ Yl Ω(AB|C) (ρ) . the 1-dimensional argument based on MPS in Sec.(V),
(3.9) and the general argument based on symmetric low-depth
circuits in Sec.(VI), the fact that the symmetry under
Using these equations for different charges κ ∈ Q, and consideration is Abelian is crucial. It is interesting to
(A) (B) see how in these arguments, which are completely dif-
different local operators Xk and Yl , one can re-
construct state Ω(AB|C) (ρ), and hence find the SPT- ferent from each other, this assumption is needed in dif-
Entanglement of ρ, between A and B relative to C. ferent ways. For instance, in the argument in terms of
(A) symmetric low-depth circuits, we use the fact that the
In particular, suppose the local operators {Xk }k and
(B)
charges corresponding to an Abelian group are additive.
{Yl }l form an orthonormal basis for the space of lo- That is the total charge in one region can be written as
cal operators acting on A or B, respectively, such that the sum of the charges in its subregions. On the other
(A) (A) † (B) (B) †
Tr(Xk Xk0 ) = δk,k0 and Tr(Yl Yl0 ) = δl,l0 . Then, hand, in the case of 1-dimensional systems, where the
from Eq.(3.9) we find SPT phases can be classified in terms of the cohomol-
ogy class of projective representations of the symmetry,
XX (A) † (B) †
Ω(AB|C) (ρ) = s̃k,l (κ) |κihκ|(KC ) ⊗ Xk ⊗ Yl . to prove our result we use another interesting fact about
κ k,l
Abelian symmetries: All irreducible projective represen-
(3.10) tations of an Abelian group in the same cohomology class
have the same dimension (See lemma 5).
Therefore, choosing any measure of entanglement, we can Of course the notion of charge measurement can be
quantify the SPT-Entanglement of state in terms of the extended to the case of non-Abelian groups, where the

charge can be interpreted as the label for different irreps
of the group. However, it can be easily shown that for
non-Abelian groups, if we measure the charge in region
C, then regions A and B can become entangled, even if
the system is in the trivial phase (Note that unlike the
case of Abelian charges, to measure a non-Abelian charge
in region C we need to act on all the sites in this region
collectively, and this can create entanglement between
A and B). It follows that for non-Abelian charges the
SPT-entanglement is not a universal quantity.
We remind the reader that the Abelian group in the
definition of SPT-Entanglement could be a subgroup of
a possibly non-Abelian group that protects the phase.
2. Entanglement versus correlation
Whenever one uses entanglement measures to detect
or understand a physical phenomenon in the context of
many-body systems, one should ask if entanglement is
really the relevant property to consider. In other words,
why we should look at a measure of entanglement and not
other measures of correlations such as mutual informa-
tion, which are often easier to calculate (Note that in the
special case of bipartite pure states, mutual information
is just twice the entanglement entropy, and therefore is a
measure of entanglement. In other words, in this special
case entanglement and correlation can be thought to be
the same properties).
As we discussed in Sec.(II C) the defining property of
measures of entanglement which distinguishes them from
measures of correlations, is their monotonicity under clas-
sical communication. But, in the context of many-body
systems or field theories it is not often clear that why
this distinction between classical and quantum commu-
nication is relevant at all.
Interestingly, the monotonicity of measures of entan-
glement under classical communication plays a clear role
in both arguments that we present to prove the properties
of SPT-Entanglement (i.e. the general argument based
on low-depth circuit in Sec.(VI), and the 1-dimensional
argument in Sec.(V)). One of the reasons that mono-
tonicity of entanglement measures under classical com-
munication is important in these arguments is the follow-
ing fact: When we measure the charge in region C then
regions A and B can become correlated. In particular, if
the total charge in A, B and C is known, then measuring
the charge in region C can induce correlations between
the charges in region A and region B (They should sum
up to a known value, and therefore they should be cor-
related). But this is a classical correlation between the
charge degrees of freedom of regions A and B. There-
fore, using measures of entanglement we can automati-
cally filter out this kind of classical correlations, which
are induced by the measurement in region C.
9
3. Total charge or the charges of all the individual sites?
In the definition of state Ω(AB|C) (ρ), the register KC
only keeps the information about the total charge in re-
gion C. In other words, Alice and Bob do not know the
charges of the individual sites in this region. But, what
if instead of the total charge, they are given the informa-
tion about the charges of all the individual sites in region
C? Does the entanglement between A and B remain uni-
versal if Alice and Bob have this extra information? One
can easily construct examples to show that the answer is
no. In other words, having this extra information about
the charges in region C, in some cases allow Alice and
Bob to extract more entanglement from regions A and
B.48
4. Topology of the three regions
Another assumption which is made in the definition
is that, region C is surrounded by regions A, B and the
boundaries of the system. This condition guarantees that
if we apply a local symmetric unitary to the system, then
the charge in this region either goes to region A, or region
B. Therefore, having access to these regions one can find
the original charge in region C. Again, one can easily
construct examples to show that if this condition does not
hold, then the SPT-Entanglement will not be a universal
quantity.
IV. EXAMPLE: PERTURBED CLUSTER
HAMILTONIAN
The cluster Hamiltonian for qubits (spin-half systems)
on a ring is defined by
X X
Hclus = − Zi−1 Xi Zi+1 = − Ki , (4.1)
i i
where X and Z denote the Pauli σx and σz operators,
and Ki = Zi−1 Xi Zi+1 . For simplicity we assume the
ring has even number of qubits (The Hamiltonian can
also be defined on an open chain, where one can add two
commuting boundary terms to remove the degeneracy
(See e.g.42 )). It can be easily seen that all the terms Ki
commute with each other, and their eigenvalues are ±1.
Hence the Hamiltonian is gapped, and exactly solvable.
Indeed, using the standard results in the stabilizer for-
malism, it can be easily shown that Hclus has a unique
ground state |Ψclus i, called the (1-dimensional) cluster
state, which is the common eigenvector of all Ki opera-
tors with eigenvalue 1, that is
Ki |Ψclus i = |Ψclus i . (4.2)
This state plays an important role in the Measurement-
Based Quantum Computation (MBQC)43 . The cluster
 10

Hamiltonian and its perturbed versions have been ex- representations. These charges can be labeled by two
tensively studied in the recent years. In particular, for bits ro re ∈ {00, 01, 10, 11}, such that the group element
the case where the Hamiltonian is perturbed byP uniform bo be ∈ {00, 01, 10, 11} is represented by (−1)ro bo +re be .
magnetic field in x direction, i.e. by the term B i Xi , it Let C be a connected region with even number of
has been shown that the Hamiltonian has quantum phase qubits, which could be arbitrary large. Then, the projec-
transition at |B| = 142,45 . tor to the subspace with charge labeled by ro re in region
NThe cluster Hamiltonian
N commutes with the unitaries C is given by
X
i∈even i and i∈odd i that is all Pauli x operators
X , (C) (C)
acting on even qubits or odd qubits, respectively. These I + (−1)ro Xodd
I + (−1)re Xeven

Π(C)
ro re = , (4.3)
unitaries clearly form a representation of the group Z2 × 2 2
Z2 . By grouping pairs of neighbor qubits together, we
where
can interpret each pair as a site. Then, for any pair of
Y Y
qubits, unitaries X ⊗ I and I ⊗ X generate an on-site (C)
Xeven = Xi , and
(C)
Xodd = (4.4)
representation of Z2 × Z2 11,12 (See Sec.(II A 1)). i∈C i∈C
As we discussed before, according to the classification i:even i:odd
of SPT phases in 1-dimensional spin systems, there is are the product of X operators on even and odd qubits
only one non-trivial SPT phase protected by Z2 × Z2 in region C, respectively. In other words, measuring the
symmetry, and it has been shown that the cluster states total charge in region C is equivalent to measuring both
is in this non-trivial phase11,12 . (Note that in the absence (C) (C)
operators Xeven and Xodd . Therefore, the charge mea-
of the symmetry, the cluster state can be transformed to
surement can be done, for instance, by measuring X op-
a product state via a low-depth circuit, and therefore
erators on all qubits in region C, and then looking to
it does not have topological order). Remarkably, this
the total parity of the outcomes of these measurements
property of cluster state has been linked to the fact that
for odd qubits (which determines the bit ro ), and even
it is a computational resource in the context of MBQC,
qubits (which determines the bit re ). Incidentally, this is
and more specifically, to the fact that it can be used as a
how the identity gate is performed on the 1-dimensional
quantum wire 11,12 . The computational power of cluster
cluster state in the MBQC (See Sec.(IV C) for further
state is also closely related to the fact that it has a non-
discussion).
zero SPT-entanglement.
Suppose C is surrounded by two regions A and B in
We have seen that in the case of 1-dimensional sys-
the left side and the right side respectively, where each
tems and Abelian symmetries, the SPT-Entanglement
region contains at least two qubits (See Fig.(2)). Let
of state is determined by d[ω] , the dimension of the ir- (AB)
reducible projective representations in the cohomology |Φro re i be the post-measurement joint state of regions
class [ω] corresponding to the SPT phase of the system. A and B corresponding to the charge labeled by ro re in
In the case of group Z2 × Z2 , the dimension of these region C. Then, for the case of cluster state Ψclus , the
irreducible representations are either one or two, corre- state Ω(AB|C) (Ψclus ) defined in Eq.(3.1) is given by
sponding to the trivial and non-trivial phase (The set 1
X
of Pauli operators together with the identity operator pro re |ro re ihro re |KC ⊗ trAB (Π(C)
r0 r1 |Ψclus ihΨclus |) ,
forms an irreducible projective representation of Z2 × Z2 ro ,re =0
with dimension two). Therefore, for the non-trivial SPT (4.5)
phase protected by Z2 × Z2 , we have d[ω] = 2. Hence,
according to our general results, the SPT-Entanglement where pro re is the probability of outcome ro re in the
of 1-dimensional cluster state should be equal to the en- charge measurement.
tanglement of a maximally entangled state of a pair of Then, using the nice properties of the cluster state,
2-dimensional systems. one can easily show that (i) each of these four charges
In the following, we examine this claim in more de- are obtained with equal probability 1/4, and (ii) for each
tails. Indeed, we show that the claim holds true for the particular charge ro re ∈ {00, 01, 10, 11}, region A is en-
grounds states of a general class of Hamiltonians obtained tangled with region B and this entanglement is equal to
by perturbing the cluster Hamiltonian. Note that this the entanglement of a maximally entangled state of a pair
result can be interpreted as an independent proof of the of qubits (A general proof is presented below). More pre-
fact that the cluster state is in the non-trivial SPT phase cisely, assuming the outcome of the charge measurement
protected by Z2 × Z2 . in region C is ro re , then up to local unitaries on A and
B, the joint state of these regions is given by
ro re
A. SPT-Entanglement for cluster state |Φ(AB)
ro re i = (Ia ⊗ Xb Zb Hb )|Θab i ⊗ |Φ
Arest
i|ΦBrest i ,
(4.6)
where
To calculate SPT-Entanglement, first recall from
Sec.(II A 1) that the group Z2 × Z2 has four possible |00iab + |11iab
|Θab i = √ (4.7)
charges, corresponding to four 1-dimensional irreducible 2
 11

is a maximally entangled state of a qubit a in region A B. A general argument based on the symmetries of
and a qubit b in region B, |ΦArest i is the state of the rest the Hamiltonian
of qubits in region A, |ΦBrest i is the state of the rest of
qubits in region B, and Hb is the Hadamard operator Let C be a connected region on the ring. For simplicity
acting on qubit b. assume C has even number of qubits. Let qubits a and
Therefore, we conclude that in this example state a0 be, respectively, the first and the second qubits imme-
Ω(AB|C) (Ψclus ) defined in Eq.(3.1) is equal to diately outside C in the left-hand side of this region, and
qubits b and b0 be, respectively, the first and the second
1
1 X qubits immediately outside C in the right-hand side of
|ro re ihro re |KC ⊗ trAB (Π(C)
r0 r1 |Ψclus ihΨclus |) this region. Define the unitaries
4 r ,r =0
o e

1 (C) (C)
1 X Fodd = (Ia0 ⊗ Za ) ⊗ Xodd ⊗ (Xb ⊗ Zb0 ) , (4.12a)
∼ |ro re ihro re |KC ⊗ |Φ(AB) (AB)
= ro re ihΦro re | , (4.8) (C) (C)
4 r ,r =0 Feven = (Za0 ⊗ Xa ) ⊗ Xeven ⊗ (Zb ⊗ Ib0 ) , (4.12b)
o e

where ∼= means the equality holds up to local unitaries where I, X, Y , and Z are, respectively, the qubit identity
on systems A and B, which do not change the entangle- operator, and Pauli operators. Note that
ment. Next, note that because state |ΦArest i|ΦBrest i is (C) (C) (C) (C)
uncorrelated across A and B, the entanglement of state [Feven , Fodd ] = [Fodd , Hclus ] = [Feven , Hclus ] = 0 .
Ω(AB|C) (Ψclus ) relative to A|KC B partition, or AKC |B (4.13)
(C) (C)
partition, is equal to the entanglement of state These equations follow from the fact that Feven and Fodd
can be written as products of Ki operators, namely
1 h Y i h Y i
1 X
|ro re ihro re |KC ⊗(Ia ⊗Uro re )|ΘihΘ|ab (Ia ⊗Ur†o re ) (C)
Feven = Ka Ki , and Fodd =
(C)
Ki Kb .
4 r ,r =0
o e i∈C i∈C
(4.9) i:even i:odd

where Uro re = Xbro Zbre Hb . As we have seen in (4.14)

Sec.(II C 1), this state can be reversibly transformed to
state |Θiab via LOCC, and hence for any measure of en-
tanglement E, its entanglement relative to A|KC B par-
tition or AKC |B partition is equal to the entanglement
of state |Θab i, i.e.
E(Ω(AB|C) (Ψclus )) = E(|Θab i).
For instance, if we use the negativity N (σ AB ) ≡
kσ AB
TA
k1 −1
2 , as a measure of entanglement, then the
above equation turns out to the following form
1
h iTA
degenerate eigenstate of this Hamiltonian, by measuring
1 X
tr (Π(C)
−1=1 . (C) (C)
AB r0 r1 |Ψ clus ihΨ clus |)
the observables Xeven and Xodd , and performing a uni-
2 r r =0 1 2 2
0 1
(4.11)
Note that this equality holds regardless of the size of
regions A, B and C (as long as they contain at least two
qubits).
In conclusion, we see that this example is consistent
with our general results: For the non-trivial phase of
Z2 × Z2 , the dimension of the irreducible projective
representations is d[ω] = 2, and therefore the SPT-
Entanglement should be equal to the entanglement of
a maximally entangled state of a pair of qubits, which is
exactly the case for the cluster state.
In the following, we present a general argument to show
that why after measuring the charge in region C on the
cluster state, the regions A and B should become entan-
gled. This argument is based on some symmetries of the
cluster Hamiltonian, and hence applies to a more general
class of perturbed cluster Hamiltonians.
Then, because all Ki operators commute with each other,
P (C)
it follows that Hclus = − i Ki commutes with Fodd and
(C)
Feven , and therefore these unitaries are symmetries of the
cluster Hamiltonian.
As we show in Sec.(IV D), all Hamiltonians with such
(4.10) symmetries behave similar to the cluster Hamiltonian in
the following sense:
Proposition 3 Suppose a Hamiltonian H is invariant
(C) (C)
under unitaries Feven and Fodd defined in Eqs.(4.12),
(C) (C)
i.e. [H, Feven ] = [H, Fodd ] = 0. Then, given any non-
tary transformation on qubits aa0 and a unitary trans-
formation on bb0 we can transform qubits a and b to a
maximally entangled state.
More precisely, we show that given a non-degenerate
(C)
eigenstate |Ψi of Hamiltonian H, if the outcome of Xeven
re
measurement is (−1) for re ∈ {0, 1} and the outcome
(C)
of Xodd measurement is (−1)ro for ro ∈ {0, 1}, then by
applying a Controlled-Z unitary (defined below) on aa0
and a Controlled-Z unitary on bb0 we can transform the
post-measurement state of system to
|Φro re i = (Ia ⊗ Xbro +fo Zbre +fe Hb )|Θab i ⊗ |Φab i , (4.15)
where |Φab i is the state of all qubits except a and b,
and Hb is the Hadamard operator acting on qubit b.
Furthermore, the two bits fo , fe ∈ {0, 1} are defined by
(C) (C)
(−1)fe = hΨ|Feven |Ψi and (−1)fo = hΨ|Fodd |Ψi. Finally,

the controlled-Z unitary is the two-qubit unitary which
maps |11i to −|11i and leaves states |00i, |01i, and |10i
unchanged.
Note that state in Eq.(4.6), which is obtained from the
cluster state, is a special case of state in Eq.(4.15)(For
cluster state fe = fo = 0).
Proposition 3 has the following corollary: Suppose we
add a perturbation to the cluster Hamiltonian Hclus ,
which (i) acts trivially on qubits a, a0 , b, and b0 , and
(C) (C)
(ii) commutes with Xeven and Xodd . Then, if an eigen-
state of the perturbed Hamiltonian is non-degenerate, by
(C) (C)
measuring Xeven and Xodd , and applying local unitaries
0
on qubits aa0 and bb , we can transform qubits ab to a
maximally entangled state.
This corollary implies that the SPT-Entanglement for
non-degenerate eigenstates of such Hamiltonians is non-
zero regardless of the size of region C, and therefore, it
follows from our results on SPT-Entanglement that the
state should be in the non-trivial SPT phase. That is it
cannot be transformed to a product state by a low-depth
circuit
N which respects
N Z2 × Z2 symmetry represented by
X
i∈even i and i∈odd Xi .
For instance, suppose we perturb the cluster Hamilto-
nian by adding a (possibly non-uniform) magnetic field
in x direction everywhere throughout the system, except
on the four qubits aa0 , and bb0 , where the distance be-
tween aa0 and bb0 is arbitrary large, i.e. of the order
of the system size. Then, our result implies that if the
ground state of such Hamiltonian is non-degenerate, then
it should be in the non-trivial SPT phase. More specif-
(C)
ically, we know that by measuring operators Xeven and
(C)
Xodd and applying controlled-Z unitaries on aa0 and bb0 ,
we can transform qubits ab to a maximally entangled
state, similar to the case of cluster state.
Note that the maximally entangled state we obtain in
this way might be different from the one we obtain from
the cluster state. This is because the final state of a and
(C) (C)
b, not only depends on the outcomes of Xeven and Xodd
measurements (encoded in ro and re ), but also depends
on the value of the bits fe and fo . This fact has important
consequences in the context of MBQC. We discuss more
about this in Sec.(IV C)
In the above argument we assumed the perturbation
acts trivially on the four qubits aa0 , and bb0 . Clearly
this assumption does not hold if, for example, we apply
a uniform magnetic field in x direction
P to all qubits in the
system, that is if we add the term B i Xi to Hclus . But,
in this situation we can decompose the perturbation to
two parts: a part which acts trivially on these four qubits,
and a small part which acts non-trivially on them. In
the example of uniform magnetic field in x direction, for
instance, we can write
X
Hclus (B) = Hclus + B Xi (4.16a)
i unique.
= H̃clus + B(Xa + Xa0 + Xb + Xb0 ) , (4.16b)
P
where H̃clus = Hclus + B i6=a,a0 ,b,b0 Xi . Clearly, H̃clus
12
satisfies the symmetry conditions required by the above
proposition. Then, it follows form the proposition that
if the ground state of H̃clus is unique then it should have
a nonzero SPT-Entanglement, regardless of the size of
region C, and therefore, using our general results on SPT-
Entanglement, we conclude that it should be in the non-
trivial SPT phase.
Now we can think of the term B(Xa + Xa0 + Xb + Xb0 )
as a small (norm-bounded) symmetric perturbation on
the Hamiltonian H̃clus . Assume we know the ground
state of Hclus (B) is non-degenerate, and its energy gap
with the first excited state is ∆E(B). Then, if ∆E(B) >
8|B|, we can conclude that the ground state of Hamil-
tonian H̃clus is also non-degenerate, and has an energy
gap larger than ∆E(B) − 8|B| (This follows from the
fact that adding a perturbation V to a Hamiltonian can
change the energy of each eigenstate by, at most, kV k,
where k · k is the largest singular value of the opera-
tor). Indeed, considering the family of Hamiltonians
H̃clus +sB(Xa +Xa0 +Xb +Xb0 ) for s ∈ [0, 1], we can eas-
ily see that if 8|B| < ∆E(B), then there exists a smooth
path in the space of gapped symmetric local Hamiltoni-
ans which connects H̃clus to Hclus (B). This implies that
the ground state of H̃clus and Hclus (B) should be in the
same SPT phase. More precisely, the ground state of the
Hamiltonian H̃clus + B(Xa + Xa0 + Xb + Xb0 ) can be
obtained from the ground state of H̃clus via a low-depth
symmetric circuit acting in a neighborhood around aa0
and bb0 .
To summarize, we conclude that if Hclus (B) has a
unique ground state with gap ∆E(B) > 8|B| then it
should be in the non-trivial SPT phase protected by
Z2 × Z2 . This is consistent with the previously known
result in42 .
We can easily
P generalize this argument for a general
perturbation j Vj . The idea is that we choose a pair of
neighbor qubits aa0 and another pair of neighbor qubits
bb0 which are arbitrary far from each other. Then, we
adiabatically turn off all the terms Vj which act non-
trivially on each of the qubits a, a0 , b and b0 . We also
turn off all the terms which act non-trivially on both sides
of aa0 , i.e. those which connect the qubits on the left side
of aa0 to the ones on the right side of aa0 . Similarly, we
turn of all the interactions which act on both sides of bb0 .
P
Then, if the perturbation j Vj is local and the gap of
P
Hamiltonian H = Hclus + j Vj is large enough com-
pared to the norm of Vj operators, as we turn off these
interactions the gap does not vanish. This means that
the modified Hamiltonian H̃Pis connected to the original
Hamiltonian H = Hclus + j Vj via a smooth path of
gapped symmetric Hamiltonians, and therefore they are
in the same SPT phase. Furthermore, if the ground state
of H is unique then the ground state of H̃ should also be
Then, using N the facts that N
(i) all Vj operators commute
with both i∈even Xi and i∈odd Xi , and (ii) all the
remaining Vj operators in H̃ acts trivially on aa0 and b0 b,

and they do not connect the two sides of these qubits, we
(C) (C)
can easily see that [H̃, Feven ] = [H̃, Fodd ] = 0, where C
0 0
is the region between aa and b b, and for simplicity we
assume it has even number of qubits.
(C) (C)
Therefore, because [H̃, Feven ] = [H̃, Fodd ] = 0, and H̃
has a unique ground state, by proposition 3 we know that
by measuring charge in region C, one can create a max-
imally entangled state between a and b, which implies
the state should be in the non-trivial SPT phase. Since
H and H̃ are in the same SPT phase, this implies the
ground state of the original Hamiltonian H should also
be in the non-trivial SPT phase.
Using this argument we can easily prove the following
result:
Theorem 4 Consider the perturbed cluster Hamiltonian
P ter state are basically equivalent to a charge measure-
H = Hclus + j Vj , where each term Vj has Z2 × Z2
N ment: to implement the identity gate one measures X
symmetry
N (i.e. commutes with both i∈even Xi and
X ), and is k-local (i.e. it acts non-trivially,
i∈odd i
on at most k neighbor qubits). Also, assume there are,
at most, t different terms Vj which act non-trivially on
any given qubit in the system. Then, if the ground
state of H is non-degenerate and has the spectral gap
∆E > 4t(k + 1) × maxj kVj k, then it is in the non-trivial
SPT phase protected by Z2 × Z2 symmetry.
We can also directly argue that for the ground of any
such Hamiltonian, the SPT-Entanglement between two
sufficiently large regions A and B should be nonzero, re-
gardless of the distance between A and B: Suppose we
choose the connected region A such that qubits aa0 are
somewhere in the middle of region A, far from its bound-
aries. Similarly, we choose the connected region B such
that bb0 are in this region, and far from its boundaries.
Then, from 3 we know that if in the ground state of H̃,
we measure the charge in the region between qubits aa0
and bb0 then the entanglement between aa0 and bb0 is
nonzero.
Now as we add the extra terms to H̃, which act non-
trivially on aa0 and bb0 , because the system is gapped,
the effect of these local perturbations on the ground state
is equivalent to two local (symmetric) unitaries which act
in a neighborhood around aa0 and bb0 . In other words,
there are local unitaries VA and VB which act respec-
tively, in a neighborhood around aa0 and bb0 , such that
VA VB transforms the ground state of H̃ to the ground
state of H. But, if regions A and B are large enough then
the support of these unitaries are contained in these re-
gions, and therefore applying these local unitaries cannot
change the SPT-Entanglement of state between regions
A and B. It follows that in the limit where regions A
and B are large then the SPT-Entanglement should be
nonzero, regardless of the distance between A and B.
13
C. Identity gate in MBQC is the charge
measurement
As we mentioned before, it has been shown that the
computational power of the 1-dimensional cluster state
can be understood as a consequence of the fact that
this state lies in the non-trivial SPT phase protected
by the Z2 × Z2 symmetry. More precisely, it has been
argued by Else et. al11,12 that for any other state of
a 1-dimensional chain that lies in the same SPT phase,
the operation which corresponds to the identity gate in
MBQC can be implemented over arbitrary long distances
(See also14–16 ). Their argument is based on the classifi-
cation of the SPT phases in the MPS framework, and
holds for a more general class of SPT phases, which cor-
respond to the maximally non-commutative cohomology
classes11,12 .
It is interesting to note that the measurements required
to implement the identity gate on the 1-dimensional clus-
operators on all qubits between the two endpoints, and
then finds the parity of the outcomes of measurements
on the even and odd qubits. As we have seen before, this
process basically measures the total charge correspond-
ing to Z2 × Z2 symmetry in the region between the two
qubits. This simple observation may, to some extent, de-
mystify the robustness of the computational power of 1-
dimensional cluster state under symmetric perturbations,
and provide a new insight into this important result. For
instance, consider the cluster state on an open chain, and
suppose we apply an arbitrary symmetric unitary which
acts trivially on the two qubits at the two endpoints of
the chain. Under any such unitary the total charge in
the region between the two endpoints remains conserved.
This immediately implies that the transformed state can
also be used to implement the identity gate between the
two endpoints, exactly in the same way that the cluster
state itself can be used.
Finally, it is worth noting that the fact that the state is
in the non-trivial SPT phase alone, does not mean that
we can use it as a resource for quantum computation
or teleportation. For example, consider the family of
Hamiltonians
Hclus (θ) = S(θ)Hclus S † (θ) , (4.17)
which are obtained from the cluster Hamiltonian by ap-
plying the local unitary
S(θ) = cos θI + i sin θ(Zj−1 ⊗ Zj+1 ) , (4.18)
where j is an arbitrary qubit in the system.N Observe
that Nthe unitary S(θ) commutes with both i∈even Xi
and i∈odd Xi , and therefore all the Hamiltonians in
this family have Z2 × Z2 symmetry. Furthermore, all
these Hamiltonians have the same spectrum as the clus-
ter Hamiltonian Hclus , and hence they are gapped. It
follows that for any value of θ there is a smooth path of
 14

(C)
symmetric gapped Hamiltonians which connects Hclus (θ) This immediately implies that the vectors Πro re |Ψi are
to Hclus , and therefore their ground states are in the (C) (C)
also eigenstates of Fodd and Feven with eigenvalues ±1,
same SPT phase (Equivalently, their ground states can
(C) (C)
be transformed to each other via a low-depth symmetric Fodd Π(C) (C) fo (C)
ro re |Ψi = Πro re Fodd |Ψi = (−1) Πro re |Ψi
circuit, namely S(θ) itself). (4.21a)
Suppose we want to use the ground state of Hclus (θ) (C) (C)
Feven Πro re |Ψi = Π(C) (C) fe (C)
ro re Feven |Ψi = (−1) Πro re |Ψi .
to perform the identity gate between qubit j and another
qubit k 6= j ± 1 in the system. It can be easily shown (4.21b)
that if we follow the same steps that we do in the case of (C) (C) (C)
Next, observe that Πro re Xodd = (−1)ro Πro re , and
cluster state, i.e. we measure the charge in the region be- (C) (C) (C)
tween the two qubits and then decouple them from the Πro re Xeven = (−1)re Πro re , which can be seen using
rest of qubits by applying controlled-Z unitaries, then Eq.(4.20). Therefore, multiplying both sides of
we can still create a maximally entangled state between (C) (C)
Fodd = (Ia0 ⊗ Za ) ⊗ Xodd ⊗ (Xb ⊗ Zb0 )
qubits j and any arbitrary qubit k 6= j ± 1, regardless of
(C) (C)
their distance. However, the maximally entangled state Feven = (Za0 ⊗ Xa ) ⊗ Xeven ⊗ (Zb ⊗ Ib0 ) ,
we create in this case is different from the one we ob- (C)
tain in the case of cluster state, by a local unitary which in Πro re we find
depends on θ. It follows that if we do not have any infor- (C)
Fodd Π(C) ro (C)
ro re = (−1) (Za ⊗ Xb ⊗ Zb0 )Πro re , (4.22a)
mation about θ, then the average state will be the totally
(C) (C) re
mixed state, and therefore it cannot be used as a resource Feven Πr o r e = (−1) (Za0 ⊗ Xa ⊗ Zb )Π(C)
ro re . (4.22b)
for teleportation, or quantum computation. This together with Eqs.(4.21) imply that
We conclude that to be able to exploit the computa-
tional power of the perturbed cluster Hamiltonian just [Za ⊗ (Xb ⊗ Zb0 )] Π(C)
ro re |Ψi = (−1)
r0 +fo (C)
Πro re |Ψi ,
knowing that the ground state is in the non-trivial SPT (4.23a)
phase is not enough, and one needs to have further infor-
[(Za0 ⊗ Xa ) ⊗ Zb ] Π(C)
ro re |Ψi = (−1)
re +fe (C)
Πro re |Ψi .
mation about the perturbation, or equivalently about the
ground state of the perturbed Hamiltonian. In particu- (4.23b)
lar, if one assumes the perturbation is sufficiently weaker Let CZ be the controlled-Z unitary, i.e. the two-qubit
than the gap of the cluster Hamiltonian, then the maxi- unitary which maps |11i to −|11i and leaves states |00i,
mally entangled state obtained from the ground state of |01i, and |10i unchanged. Define
the perturbed Hamiltonian is close to the one obtained
(C)
from the cluster state, even if the two qubits are arbi- [CZaa0 ⊗ CZbb0 ]Πro re |Ψi
|Φro re i = q , (4.24)
trary far from each other (In the above example, this (C)
means that we need to assume |θ|  1). hΨ|Πro re |Ψi
to be the state of system after projecting region C to
the sector with charge ro re , and applying a controlled-Z
D. Proof of proposition 3 unitary on qubits a and a0 and a controlled-Z unitary to
qubits b and b0 . Then, by applying the operator CZaa0 ⊗
In the rest of this section we present the proof of propo- CZbb0 on both sides of Eqs.(4.23), and using the fact that
sition 3. CZ(X ⊗ Z)CZ = (X ⊗ I), we find
Let |Ψi be a non-degenerate eigenstate of H. The fact (Za ⊗ Xb ) |Φro re i = (−1)ro +fo |Φro re i , (4.25a)
(C) (C)
that Fodd and Feven commute with H, together with the (Xa ⊗ Zb ) |Φro re i = (−1) re +fe
|Φro re i . (4.25b)
fact that this eigenstate is non-degenerate implies that
(C) (C) It can be easily shown that these two equations together
|Ψi should also be an eigenstate of Fodd and Feven . Using
(C) (C) uniquely determine the state of qubits a and b. More
the facts that Fodd and Feven are both unitary operators,
precisely, up to a global phase, the only state of qubits a
and the square of both is the identity, we find that
and b which satisfies these equations is the state
(C)
Fodd |Ψi = (−1)fo |Ψi , (4.19a) (Ia ⊗ Xbro +fo Zbre +fe Hb )|Θab i , (4.26)
(C) fe
Feven |Ψi = (−1) |Ψi , (4.19b) where |Θab i = |00iab√+|11i ab
, and Hb is the Hadamard
2
gate. It follows that any state |Φro re i which satisfies
where fo , fe ∈ {0, 1}.
(C) (C)
these equations should be in the form of
Next, note that unitaries Fodd and Feven commute with
(C) (C)
Xeven and Xodd , and hence they commute with |Φro re i = (Ia ⊗ Xbro +fo Zbre +fe Hb )|Θab i ⊗ |Φab i , (4.27)
where |Φab i is the state of the rest of qubits in the system,
(C) (C) which is arbitrary (i.e. it is not constrained by these
I + (−1)ro Xodd
I + (−1)re Xeven

Π(C)
ro re = . (4.20) equations). This completes the proof of the proposition.
2 2
 A A BB
g2G == AA+
+ BB
⌦(AB|C) ( ) up
(+) T (+) (+)
But, since
But,(AsinceB)|⇥
(A B)|⇥ ⇧ =(+)
(AB
⇧ = (ABI)|⇥T
I)|⇥⇧, where BT BT
⇧, where
is the transpose of B we find
on systems a0 and Za and obtains charge A . Similarly,
is the transpose of B wethat
find that
⌅ ⇥ † ⇧⇥
Bob performs the local projectiveA measurement corre-
⌅ † ⇧
sponding to the projectors 15
T † u( ) (g)T
T † u( u)(⇥) u(⇥) (g)
(g)T(g) I |⇥ (+) ⇧=
I |⇥ (+) i⇤(g) i⇤(g)
e⇧ = e |⇥ (+)|⇥⇧(+) ⇧
In the follo
(0.26)
(0.26) X
V. SPT-ENTANGLEMENT that IN
This implies
This implies † (1-D
Tthat ) (⇥) † (⇥) †
† ( SYSTEMS
u )T(g)Tu u (g)T(g)
u =(g) e = e Ii⇤(g)
i⇤(g)
and so so ⇧
I and
˜ (b0 Zb ) =aL1aLaRaR(a(a
e++i(g)
1)1) iR bbLL ⇤b(g)
iVL i0 R
LLiL b (g)⌦u R :  2 Q , (6.17) the special ca
|G| V (g) of group
g2G
( ) i⇤(g) (⇥) †
⌅g ⇤ G⌅g
: ⇤
u(G) (g) e i⇤(g)
: u = (g) = e T u(⇥)T(g)T †
u (g)T (0.27)
(0.27) S S S S representation
In this section, we use the classification of the SPT be extended t
phases in the Matrix Product State (MPS) framework to i the following
calculate SPT-Entanglement in the case of 1-dimensional on systems b0 and Zb , and obtains charge B . Finally,
systems. We start with a short review of this classifica- FIG. 5: Isometry S maps virtual subsystems iL and iR to the Lemma 13 L
they add the value of the charges they found in the mea- representation
tion (we follow the presentation of37 and31 ). physical site i.
surements, and obtain the total charge  = A + B . It 1-dimensiona
1
P
can be easily shown that this process is basically equiv- ge
i(g)
0 0 |G|
˜ (a b )
alent to implementing the charge measurement {⇧ : projector |⇥(
A. MPS representation of SPT phases  2 Q}. The key point is that because of Eq.(6.15) we mally entangl
group,knowwhich
thatisstate
called
|⇥iZthe second cohomology class of G,
Z = |⇥iZa Zb has charge zero with
2 a b
and isrespect
denoted byrepresentation
to the H (G, U (1)).g ! u(g) ⌦ u⇤ (g). There- This lemma f
For a 1-dimensional system with short range correla- According to systems
the classifications
Za and Zb in of the|⇥i
SPT to sys- of is a (linear) u
fore, adding state Za Zb phases
tions by blocking the sites in the large blocks a transla- tems a0 andsystems,
1-dimensional b0 does not thechange their total
equivalence charge.
class How-
[ω] uniquely the group is A
tionally invariant ground state will converge to a fixed determines the SPT phase . For example, the 2- composed to
ever, by adding systems Z1–4
a to the Alice’s side and Z b
point in the form of to the Bob’s side, the representations of symmetry G on
cocycles corresponding to the Abelian group Z2 ×Z2 have [u(g) ⌦ u⇤ (g
both sides become (linear) non-projective representation:
two different equivalence classes, which is g correspond to two
|Ψi = S ⊗N |λi⊗N , (5.1) On Alice’s side the representations ! Va⇤0 (g) ⌦ u(g),
different
and on Bob’s side the representation is g ! Vb0 (g)⌦u (g). IV where g ! e
phases protected by this symmetry. See Sec.
⇤ i
P for further discussion about the example
Therefore, because the local representation of2symme- of Z × Z 2 . Then, the fact
where |λi = k λk |ki|ki is a virtual entangled state tion, together
try on Alice and Bob’s systems are now Abelian, they
between adjacent virtual sites with Schmidt coefficients state |⇥()i s
can measure the charges locally and add these local
{λk }, and S is an isometry which maps two virtual sites charges together to find the charge in systems a 0
and follows that in
iL and iR to a physical site i (See Fig.5)38 . This isom- B. Cohomology of Abelian groups 1-dimensional
b0 . This proves proposition 12, and completes the proof
etry can indeed be thought as a Renormalization Group of |⇥i
LOCC
! (ab)
, which by proposition 11 implies
most once.
(RG) transformation on a block38 . It is shown that for Interestingly,
LOCC it turns out that for Abelian groups the It follows f
(AB|C)
|⇥i !⌦ ( ).
any gapped 1-dimensional system with a unique ground irreducible projective representations in the same coho- g ! V (g) is
state, by blocking the physical sites, the ground state can mology class have the same dimension. Indeed, in the then
X
be approximated by a state in the form of |Ψi with an Appendix we prove thatof ⌦(AB|C) ( ) LOCC! |⇥i
2. Proof (ab)
=
accuracy which is exponential in the block sizes38 . 
Consider a symmetry group G with on-site linear uni- Lemma Similar →approach
5 Lettogthe uα (g) and g → inuβthe
we used (g)previous
be two sec-
finite-
tary representation g → u(g). After blocking L sites dimensional projective
tion, we prove
irreducible
(ab) LOCC representations of an This is true b
! |⇥i, and this together with
together this symmetry will be represented by g → U (g) Abelian group G, whose(AB|C) 2-cocycles belong
LOCC to the same means that th
proposition 11 imply ⌦ ( ) ! |⇥i. them to the c
on each block, where U (g) = u(g)⊗L . Assume |Ψi is cohomology class. Then, these representations have the
invariant under this symmetry, i.e. U (g)⊗N |Ψi = |Ψi. same dimensions. Furthermore, there exists a unitary W
Then, it turns out that there always exists a projective and a phase eir(g) such that
representation g → V (g) of group G such that
  ∀g ∈ G : uβ (g) = eir(g) W uα (g)W † . (5.6)
Ui (g)S = S ViL (g) ⊗ Vi∗R (g) (5.2)
Basically, this lemma means that in the case of Abelian
and groups, up to a unitary and a phase freedom, there is only
 ∗  a unique irreducible projective representation in each co-
ViR (g) ⊗ V(i+1)L (g) |λi = |λi , (5.3) homology class. Note that this is not true for a general
non-Abelian group.
where V ∗ (g) is the complex conjugate of V (g)33 . Note In the following, the dimension of the projective irre-
that for any phase eiθ(g) the representation g → ducible representations in the cohomology [ω] is denoted
eiθ(g) V (g) will also satisfy the above equations, and hence by d[ω] .
V (g) is defined only up to a phase. Let ω be the 2-cocycle
of the representation g → V (g), i.e.
C. Main result in 1-dimensional systems
∀g, h ∈ G : V (g)V (h) = ω(g, h)V (gh) . (5.4)

Then, the gauge transformation g → eiθ(g) V (g) induces
an equivalence relation on the space of the 2-cocycles: ω
and ω 0 are equivalent, or belong to the same cohomology
class, if there exists a phase eiθ(g) for which
∀g, h ∈ G : ω(g, h) = ω 0 (g, h)ei[θ(gh)−θ(g)−θ(h)] . (5.5)
For any 2-cocycle ω, its corresponding equivalence class
is denoted by [ω]. The set of equivalence classes forms a
We calculate the SPT-Entanglement for states in the
form of state in Eq.(5.1), that is for the fixed points of
the RG. For a general MPS, which is not in this form,
we know that by blocking the sites, the state converges
to a fixed point in the form of Eq.(5.1). Therefore, for a
general MPS if we calculate SPT-Entanglement for large
regions A, B and C the result should converge to what
we obtain for these fixed points. Note that other ap-
proaches have been recently proposed for detecting SPT
 16

order based on the properties of the fixed point of the Another way to phrase this result is that, by local
RG40,41 . operations on Alice and Bob systems, the two states
Let A and B be any two non-neighbor blocks of a 1- Ω(AB|C) (Ψ) and σ (ab) can be transformed to each other.
dimensional system, and C be a (connected) region be- This implies that state Ω(AB|C) (Ψ) and state σ (ab) have
tween A and B. exactly the same entanglement properties.
(a0 b0 )
Note that the set of projectors {Π̃κ : κ ∈ Q} de-
Theorem 6 Let |Ψi be a state in the form of Eq.(5.1) scribes the charge measurement on the virtual systems
in the SPT phase corresponding to the cohomology class a0 and b0 .
[ω]. Then, with respect to any measure of entanglement This proposition follows from the following simple ob-
E, the SPT-Entanglement between A and B is equal to servations which are depicted in Fig.(6): (i) Applying
the entanglement of a maximally entangled state of a pair local isometries do not change the entanglement of state.
of d[ω] -dimensional systems. Therefore, instead of entanglement in the physical space,
In particular, if we use the negativity to quantify SPT- we can look at the entanglement in the virtual space. (ii)
Entanglement, then this theorem implies In the state Ω(AB|C) (Ψ) the only virtual system in Al-
ice’s side which can be correlated with a virtual system
i TA
X h in Bob’s side is the one at the boundary of regions A
tr (Π(C) = d[ω] ,
AB κ |ΨihΨ|) (5.7) and C, that is the virtual system aR in Fig.(6). Sim-
κ 1
ilarly, the only virtual system in Bob’s side which can
be correlated with a virtual system in Alice’s side is the
which is equal to 1 in the trivial phase, and larger than
one at the boundary of regions B and C, that is the
or equal to 2 in the non-trivial SPT phases. Note that
virtual system
 bL in Fig.(6).  (iii) Using the fact that
since the SPT phase of state |Ψi is uniquely determined
Ui (g)S = S ViL (g) ⊗ Vi∗R (g) for all g ∈ G, we find that
by the equivalence class [ω], this theorem implies that
the charge measurement in the physical space can be in-
for states in the form of Eq.(5.1), the SPT-Entanglement
terpreted as a charge measurement in the virtual space,
depends only on the SPT phase of the state.
defined by the projectors
Consider state
X 1 X −iκ(g) O
Ω(AB|C) (Ψ) = |κihκ|(KC ) ⊗trAB (Π(C) S † Π(C)
κ S = e ViL (g) ⊗ Vi∗R (g) . (5.11)
κ |ΨihΨ|) , (5.8) |G|
g∈G i∈C
κ

Furthermore, because ∀g ∈ G : Vi∗R (g) ⊗ V(i+1)L (g)|λi =

and let
|λi, the charge for each state |λi is zero. Therefore,
d[ω] any state |λi whose corresponding pair of virtual systems
1 X both live inside region C do not contribute in the total
|Θi ≡ p |kki (5.9)
d[ω] k=1 charge of this region. It follows that the total charge in
region C is just determined by the two unpaired virtual
be a maximally entangled state of a pair of d[ω] - systems at the left and the right boundaries of this region
dimensional systems. Then, another way to phrase (See Fig.6). Putting these facts together, we can easily
theorem 6 is that both transformations |Θi −−−−→
LOCC show proposition 7 (Note that a and b in state σ (ab) cor-
LOCC respond, respectively, to the virtual systems aR and bL
Ω(AB|C) (Ψ) and Ω(AB|C) (Ψ) −−−−→ |Θi can be imple-
in Fig.6).
mented via LOCC. In the following we prove this result.
Therefore, because states σ (ab) and Ω(AB|C) (Ψ) have
the same entanglement properties, to prove theorem 6 we
can focus on the entanglement properties of state σ (ab) .
D. Finding SPT-Entanglement in MPS framework

The first step in the proof of theorem 7 is the following 1.

LOCC
Proof of |Θi −−−−→ Ω(AB|C) (Ψ)
simple observation, which is depicted in Fig.(6).

Proposition 7 Applying local operations on Alice’s It follows from the above proposition that, to prove
LOCC LOCC
and Bob’s systems, we can reversibly transform state |Θi −−−−→ Ω(AB|C) (Ψ) it suffices to show |Θi −−−−→
Ω(AB|C) (Ψ) to state σ (ab) . In other words, we need to show that by consum-
X ing a copy of the maximally entangled state |Θi one can
0 0
σ (ab) = |κihκ|(KC ) ⊗ Tra0 b0 (Π̃κ(a b ) [|λihλ|aa0 ⊗ |λihλ|b0 b ]) , prepare a copy of state σ (ab) .
κ To show this first note that state σ (ab) can be generated
in the following way: Alice locally prepares state |λiaa0 ,
where and Bob prepares |λib0 b , where
0 0 1 X −iκ(g) (Va∗ (g) ⊗ Va0 (g))|λiaa0 = |λiaa0 , (5.12a)
Π̃(a
κ
b)
= e Va0 (g) ⊗ Vb∗0 (g) . (5.10)
|G| (Vb∗0 (g) ⊗ Vb (g))|λib0 b = |λib0 b , (5.12b)
g∈G
 A B A B
(+)
T T T
B⇥I)|⇥⇧, (+)
(+)where
⇧⇧,=⇧where B TBTTI)|⇥ (+)(+) BT 9
B)|⇥ B)|⇥ (+) (AB
= (AB I)|⇥ ⇧, ⇧, where
where BT 9 9
of
se B of we
B we find that
find that 9
nd (+)which ⇥ ⇥ completes
⇧ ⇧(+) the⇤proof.
g)Tu)|⇥
)and
e⇧(⇥)
=
† ⇧†i⇤(g)
i⇤(g)
u(⇥)
I)|⇥ (+)
e(g)
where
which
|⇥ |⇥
⇧ where
(g)
|⇥
I⇧ completes
(+) (+) ⇧⇧ ⇥
(+)(+)
|⇥ (+)
I|⇥⇧|⇥ =⇧⇧ ⇥
⇤1the
e d⇤
=
d
⇤
e1d i⇤(g)
i=1 d |ii⇧
proof.
i⇤(g)
|⇥ ⇧and
(+)(+)
|⇥ |ii⇧ ⇧and
which completes the proof.
which completes the proof.
17
al i=1
ytity
onal operator acting on the d-dimensional
)
26) operator acting
(0.26)(0.26) on the d-dimensional (0.26)
(0.26)
.(0.24) implies † aa
†LL aa (a
(a +
+ 1)1)L LLaiL a ai R(a(abLL+1)1)
b+ bLbRLRiL
iL aiLiRaRLaR
babLLR(a(a
I aand b+
R+
Eq.(0.24) )( implies 1)1)
L LiLiL iRiR bL
bL bRbR cohomology class have the same dimension, and therefore
so
g)
that †T
ue(† u
e T=i⇤(g) i⇤(g)
I(g)T)
⇧I ⇧uand
and sou(⇥)
(g)T(⇥) so
(g) (g)
== eRRei⇤(g)i⇤(g)
I and soLsoR R
5)
.25) (⇥) ⇥ ⇥
|⇥ (+)
⇧=u i⇤(g) (+) 9 the dimension of this representation is d[ω] . Also, note
(⇥)u)†( u)(⇥)
(g) e (⇥) A †|⇥†|⇥ ⇧ (0.25)
= A+ B
(+)
i⇤(g) i⇤(g) (+)
T
)u:() u(g)T
27)
g)T (g)(g) † (g)
= e |⇥
i⇤(g)
e=(0.27) T uT⇧(⇥)
(0.27) = e (g)T
(g)T
S A
B ⇧ (0.27)
SB (0.27) (0.25)
= A+ B S A A BB == AA++ BBS
T
that9 state |ΘiZa Zb satisfies
nd
B TB)|⇥
B)|⇥ which
(+)
⇧ =⇧ (AB
(+) completes
T
I)|⇥ the
(+) proof.
⇧, ⇤
where BT T
= (AB T(+) I)|⇥ (+) ⇧, dwhere B
nal
oseof I)|⇥
B
of we
(+)
B we ⇧ where
find that
find |⇥ a⇧ ⇥ ⇤1d a+1i=1 |ii⇧ and
that
i
which completes the proof. b
[u∗ (g) ⊗ u(g)]|ΘiZa Zb = |ΘiZa Zb . (5.13)
ntity operator
† †
⇥ ⇥ ⇧acting
⇧ A
on the d-dimensional C B
⇧Eq.(0.24)
Tg)Tu(⇥)u(⇥) (g)(g)impliesI I|⇥ (+)
|⇥ (+) e ei⇤(g)
⇧ =⇧ = i⇤(g)
|⇥|⇥(+)(+)
⇧ ⇧
⇧
This means that if we assume the representation of sym-
5)
6)
.26) (0.26)
⇥
T ) u(⇥) (g) |⇥ (+) ⇧LL=aeaRRAi⇤(g)
† a†a (a
(a |⇥
+
(+) (0.26)
1)L⇧ i=
B 1)
+ A + bb B bb
(0.25)
L iiR aLaLaRaR(a(a++1)
1) i iL iRiAR +bLbLB bRbR metry on system Za is g → u∗ (g) and on system Zb is
soT † u
othat
at T(† u)((g)T
)
u(⇥)
(g)T u(⇥)(g) (g)
== i⇤(g)
e ei⇤(g) I and
I and so so R LL RR A B L L L=
B T g → u(g), then state |ΘiZa Zb has charge zero.
(+) T (+) T
A.27) B)|⇥
( ) ⇧ e= (AB
i⇤(g) (⇥)(⇥) I)|⇥ † † ⇧, where B It follows that to perform the charge measurement
u:( u) (g)(g)
7) = e= i⇤(g) T uT u (g)T (g)T (0.27)
(0.27)
ose of B we find that (a0 b0 )
⇥ ⇧ A C B {Π̃κ : κ ∈ Q} on systems a0 and b0 , Alice and Bob
†
(g)T u(⇥) (g) I |⇥ (+) ⇧ = e i⇤(g)
|⇥ (+) ⇧ can perform the local charge measurements on systems
6)
a† aR (ai⇤(g)
(0.26)
+ 1)L iL iR bL bR
a0 Za and b0 Zb (See Fig.7). More precisely, Alice performs
aL aR (a + 1)L iL iR bL bR
othat T † u( ) (g)T u(⇥) L(g) = e I and so the measurement corresponding to the projectors
G7): u( ) (g) = e i⇤(g)
T u(⇥) (g)T † (0.27)
0 1 X −iκ(g)
A C B Π̃(a
κ
Za )
= e Va0 (g)⊗u∗ (g) : κ ∈ Q , (5.14)
|G|
g∈G
FIG. 6: Top: We look at the entanglement of regions A and
B, given the value of charge in region C is known. Here, each on systems a0 and Za and obtains charge κA . Similarly,
isometry S maps two virtual systems to a physical system. Bob performs the local projective measurement corre-
Middle: Applying local isometry S does not change entangle- sponding to the projectors
ment, or charge. Therefore, we can ignore the isometries and
look at the charge and entanglement at the level of the virtual 0 1 X −iκ(g) ∗
Π̃(b
κ
Zb )
= e Vb0 (g) ⊗ u(g) : κ ∈ Q , (5.15)
systems. Bottom: For each pair of virtual systems in state |λi |G|
g∈G
the total charge is zero. Thus, if both pairs are inside region
C then they do not contribute in the total charge in this re- on systems b0 and Zb , and obtains charge κB . Finally,
gion, and therefore we can ignore them. Furthermore, all the
they add these charges together, and obtain the total
virtual subsystems in A or B, except the ones at the bound-
ary, are uncorrelated with the virtual systems in the other charge κ = κA + κB .
regions. Therefore, since we are only interested in the entan- Then, using the fact that state |ΘiZa Zb has charge zero
glement between the two regions, we can ignore them. The with respect to the representation g → u∗ (g) ⊗ u(g) (See
resulting state is state σ (ab) defined in proposition 7, where a Eq.(5.13)), we find that adding systems Za and Zb in
and b correspond, respectively, to the virtual systems aR and state |ΘiZa Zb does not change the total charge. There-
bL in this figure. fore, the charge κ = κA + κB is equal to the total charge
in systems a0 and b0 .
This proves proposition 8, and completes the proof
LOCC
of |Θi −−−−→ σ (ab) , which by proposition 7 implies
for all g ∈ G. Then, they perform the projective measure- LOCC
(a0 b0 ) 1
P −iκ(g) |Θi −−−−→ Ω(AB|C) (Ψ).
ment {Π̃κ = |G| g∈G e Va0 (g)⊗Vb∗0 (g) : κ ∈ Q}
0 0
on systems a and b , and keep the outcome of the mea-
surement in register KC . In other words, they measure LOCC
2. Proof of Ω(AB|C) (Ψ) −−−−→ |Θi
the total charge in the systems a0 b0 . Clearly the resulting
state will be state σ (ab) . Therefore, to prove statement
LOCC In the following, similar to the approach we used in the
|Θi −−−−→ σ (ab) , it suffices to show the following LOCC
previous section, we prove σ (ab) −−−−→ |Θi, and this to-
Proposition 8 The projective charge measurement LOCC
gether with proposition 7 imply Ω(AB|C) (Ψ) −−−−→ |Θi.
(a0 b0 )
{Π̃κ : κ ∈ Q}, can be implemented via LOCC, by LOCC
First, we prove σ (ab) −−−−→ |Θi, for the special case
consuming a maximally entangled state of a pair of
where the projective representation g → V (g) of group
d[ω] - dimensional systems, where d[ω] is the dimension
G on the virtual systems is an irreducible representation,
of the irreps in the cohomology class of representation
and then explain how this argument can be extended to
g → V (g).
the general case. The proof is based on the following
To prove this, we assume Alice and Bob are given a lemma
pair of d[ω] -dimensional ancillary systems Za and Zb ,
Lemma 9 Let g → u(g) be an irreducible projective
which are prepared in the maximally entangled state
Pd[ω] representation of an Abelian group G. Then, for any
|ΘiZa Zb ≡ √ 1 k=1 |kiZa |kiZb . Let g → u(g) be an 1-dimensional g → eiκ(g) of group G,
d[ω]
1
P −iκ(g) representation
∗
irreducible representation of group G in the same coho- |G| ge [u(g) ⊗ u (g)] is either zero, or a rank one
mology class that the representation g → V (g) belongs to projector |Θ(κ)ihΘ(κ)|. Furthermore, |Θ(κ)i is a maxi-
it. From lemma 5 we know that all the irreps in the same mally entangled state.
 tion Proposition
8 0The 8 The projective
projective charge measurement charge measurement charges together chargesto together find theto tochargefind the in systems charge in a0 systems
and a00 and
Proposition (a b 0
)
˜0 b0 ) Proposition
8 The projective charge Proposition measurement 8 The charges
projective
0
b0 . This
togethercharge
provestogether
find the
measurement
proposition
chargecharges
8,together
and
in
completes
systems together a and to find the charge 0 in systems a0 and
˜{(a
 ⇧2 Q}, :can 0 2 be Q},
implemented 8Proposition
can The projective
be0 0implemented
via
8 The
LOCC, ˜ (acharge
0via
bby0 projective
) LOCC, b0measurement
. This bycharge proves
b . Thismeasurement
charges
propositionproves 8, and
proposition charges to find
completes 8, andthethe to
0chargeproof
completes find in the systems
the
proof
charge
proof a0 in and systems a and
{⇧  :  2
(a b ) 0 Q}, can be
˜ (a implemented
b) {Q}, via
⇧ can LOCC,
: abe pair 2 Q}, by can bevia implemented
0 LOCC via
(ab) bLOCC, 0
.whichThis by proves
b . This
proposition
proves proposition
8, and completes
8, and completes the proof
the proof
ng consuming
a maximally { ⇧ ˜ a maximally
:  2
entangled entangled { ⇧
Q},  entangled
can :
state of astate 
be 2 state
implemented
pair ofof a pair of viaimplemented of
LOCC, LOCC ofby |⇥i b
LOCC,
(ab) .
LOCC This! by proves , proposition by 8,
propositionand completes7 implies the proof
consuming 
a maximally consuming of
a |⇥i of
maximally of!|⇥i entangled , which ! state LOCC by
(ab)
of , aproposition
whichpair by
of
LOCC 7ofimplies
proposition |⇥i
LOCC
7 ! implies (ab)
, which by proposition 718implies
d[!] - dimensional
ensional consuming
systems, where systems, consuming
a maximally d[!]where where
is adimension
thedentangledd[!]maximally is thestate entangled
dimension of LOCC a pair state |⇥i of of
LOCC ofa!pair |⇥i ⌦[!] (AB|C)of (!of ). |⇥i
(ab)
, which ! by(ab)proposition , which by7 proposition implies 7 implies
d [!] - dimensional systems, [!] d is[!]the - dimension
dimensional |⇥i systems,
! ⌦
|⇥i
(AB|C) LOCC
where (
! ).
⌦d (AB|C) is the
( dimension
). LOCC (AB|C)
reps of in thethe irreps
d[!]cohomology in the cohomology
- dimensional dclass
[!] -systems,
dimensional
of representationclass
where ofsystems,
drepresentation
[!] is the where dimensiond[!] is the |⇥i dimension LOCC
⌦(AB|C)
! representation
LOCC
|⇥i ( ).! ⌦(AB|C) |⇥i ( ).! ⌦ ( ).
of the irreps in the cohomology class of representation
).g ! g ! V (g). of the irreps in
V (g).g ! V (g).
of the
!"
the cohomology
irreps inof the the
class irreps
cohomology
#"!"
in the
of representation class cohomologyof representation
!"
class of
#" !"
|Θ(−κ)i.
#"
#" !"
Furthermore, from this lemma we know that
be the projector g ! V (g).
charge to the subspace inbewhich
g ! V (g).
the projector
thebetotal
charge
be the
!" !"
the sys-
projector
toprojector
the charge
charge subspace to the
to the
state subspace
in subspace
|Θ(−κ)i
which thein #" intotal ab
which
which
be
is
sys-
the
a maximally
the total sys-
the total
projector
#"
sys-
to
#"
entangled state. Then,
tem formed RA Rby RA and R R be the projector using
betothe thethe fact
projector
subspace
charge alltoR maximally
in thewhich charge
subspace the inthe
entangle
total whichsubspace
sys- states
the in can
total which
sys-bethe total sys-
trans-
"# "# tem formed by charge
R and R
"! "!
tem formed B RAtem A
Rby formed B
RA R ARby R AR R and RB Aand RtemB RB A B
then after this step 0 one,0 and could , 0 v(
thenget
and rid
0then
after of this the
after step this one stepcould one get
, could
and rid | of
thenget tem
the rid
after
A formedof R
thisthe
A R by tem
step
formed R formed
one to Rcould R
eachby R
get
other formed
Band
rid R of
A
via RR Rby
the
B
local R R and
unitaries, R R and R R
a 0ato bthe , 0andb 0 then 0,after , and then
this 0 step afterone this
could
, and step
,thenget
and
v( oneridthen
after could
, and0 ofe this ,then
the
afterget
and
v( step Arid
then
after
thisone of this
step the
after
A
could step
one this
get one
A
could
rid|step
could
of get one
A
the get could
, to
ridandrid
of A
| this
of
then
B
the get
the A A
rid
after B ofthisthe B stepwe A one conclude
B could B getthat
A B of the
rid
e-h st yshc la ih taowta n ebiht beach aacpih sbb w baus aniebh ebt ca and
optasthen b rae ou
,g ts
andafter
bcra e 0jh
bthen
h octrto
this poastep
after totthe
ae one
, 0this
the
r
h
andbt,g
o
e t could
and
ce
bthen
step
r abh to
jto
one oget
,then
and
v(
after
c rthe
,the p ,rid
could
and then
afterand
this eof
,thenh this
step
get
and
v( the
after
then
t rid step
e
one
then
after after
b
of
,could
, this
toand
one
the
and
after
this step
the
this
step
then
could
step
get
,then
and
v( this one
rid
after
get
one
then
after
one of
step
tocould
,could
,and the
rid
and
the|one
could
after
thisthe
of
then get
,,then
step
get
to
step
the
get
and
v(this
could
rid
and one
(ab)
the
rid
|after
rid one
then
after
ofstep
then ofthis
of
could
get
the
could
the
this
the
after
one , step
step
ridget
after
get
this
could
of rid
and | one
one
the
this
rid
step
of
thencould
get
the
step
of
could
one the
get
rid
after
one, to get
could
ofrid
and
this
could
the
rid
|theof get
then ofafter
the
step
get the
rid
one
rid ofofthe
this
thes
could
g. 3). Note that in the state (See| d Fig.
(See 3).
Fig.Note 3). (See
Note that Fig.thatiny the
3). in state
Note
the that | rR
(See
state |in ⌦
ithe
Fig. the 3). (See
state Note |eFig.
im ⌦
that 3).tothe the
Note
infNotethe entanglement
that
state ini⌦
|in
to the
the⌦state | i ⌦ of state⌦ σ
to the in Eq.(5.17) to the is equal to the
R
To prove R
d n a
this, we R
assume n
Alice a
can begiven (See
measured
A R
Fig. 3). bR Note
via d
Fig.LOCC R e m
that A in o andf y m b R
by e
(See
state d
t |R i
Fig.⌦ r
3).o ⌦ m e
that t the state | i ⌦ ⌦ ⌦
| i ⌦
ve To this,B B
we assume A Alice B and Alice Bob areand Bobsentationsa are(See given A awhose
3). Note (See Fig.
that
can in A
3).
be the Note can
(See
(See
state
measured that be
|Fig.i in
Fig. via measured
3).
the
3).LOCC Note
Note(See
state that via
|andi inin
Fig.
that LOCC
the
3).
by theNote (See
state
stateand | Fig.
that |iby in3).
i systemsthe Note
state that| i into the state
afactor systems belong eto the iequivalence
sentations sentations whose factor belong the LOCC
equivalence
cfipair
r prove To this, prove rwe
|tancillary assume
this, To
ewe prove
assume
e fh it seand this,
fn Alice
soe eBob
dwe
To fir tare
assume
eandprovet
sne ee g given
Bob
h ithis,
t td,Alice
are
lfdoa
u weo dand
given
ttassume
c fir e nBobt oe g are
Alice
p td
e
, ⇤t lgiven
sand
u osa cih acan
Bobe
,tn robe
are
etp measured
entanglementfgiven
can a twhose
sbe
n ea tfactor
smeasured
h h via
t ofrcan
d nLOCC
e asystems
state
t fbe
via
a and
|Θi.
, measured
LOCC
n ebelong
hby tThis
and
can dn to
via
beby
a the
proves
LOCC equivalence
,belong
measured (ab)
σfactor
and
tovia −systems
by − −
LOCC −→ |Θi
and bysystems
te tne to g pd eldu
t so c st
tie
eh ,gn t
po rt
le sp t es cat n ih t r
d e
Zn t afv a tn
Ze h ,d
hn (ea
rtoa te
a e,gh dh lfuo o c in ea n o d u o h t r
o tp a ( s h h rn e a a n vh d n
dpair
[!] -dimensionalofe d[!] te
consuming h d
-dimensional o the ancillary
entangled
systems and Z[ a systems
and Ze
state
] | h tas and
,consuming
bclasses ot [ ] b ,consuming
the e
entangled t
and o[! t
the entangled
state
] and |⇥i [!
classes ⇤state |⇥i
] [!]
sentations
classess
sentations
[!]
whose factor
whose systems
sentations
factor whose
systems sentationsbelong the equivalence
to
whose belong
the factor
equivalence to the equbel
of dpair [!] -dimensional
of d [!] -dimensional ancillary
pair of d[!] -dimensional systems
ancillary pair ofZ
systemsad[!] and
ancillary Z Z
-dimensional , and systems
ab consuming Z ,
bancillary
consumingZthea and entangled
systems Zb ,consuming
the
and [!s ⇤
Z
forstate
entangled
]
and and
the
[! |⇥i
⇤ Z
state
the
]
classess, |⇥i
entangled
special
consuming and
[!]
classes case
[! ⇤ state
the
]
[!] where |⇥i
entangled g
classes
and [! → V
state
⇤[!]
] (g) |⇥i is an irreducible
| etaare
which
e prepared tsin prepared
ethe
Now ht we n | iein
maximally t ta
use athe
hts tthis maximally
e h
to
entangled tlemma
N ni .) entangled
tstate
3
a hto .tgprove iNow
eF to|eN state
ewe Sta
that.()use
t3 s the
Now .this
e g hiwe tF pro-n |e iet
use
lemma S
ta a(h
thisttost e h
to
lemma
prove
atandN nthatito . )
t 3
ab h
prove.
the tg ieF ts o
that
pro- eN e Sthe.( ) 3 pro-. gs i F e e S classes [!]
( s
which are prepared ande b Sn S use and Swe
⌘s y C
poy
s
|⇥i 1Ohb rw P
L
od which
tds ⌘
n
can
[!] aa
io
p f1are
v
|ki iC e
P tds P
aC e
do
|ki tdin
prepared
rOnh u
[!] e
wthe
which
Lss
P
.|ki
asda
Let
emaximally
nin
are
m
go
|ki !
the
iprepared
iv$%&'()"*)&+,"
te
da
u(g)
bmaximally
etwhich
. 1|ki
Let
rn Pn
uentangled
bee asa
gd.s !
an
inare
c ethe m
u(g)
state
entangled
maximally
prepared
$%&'()"*)&+,"
Pd[!]
Now state
in
a cthe entangled
weNow usewestate
maximally
$%&'()"*)&+," this useentangled
lemma
Nowthis
projective
$%&'()"*)&+,"
$%&'()"*)&+,"
lemma
we
$%&'()"*)&+," state
to prove to
Nowthisprove
$%&'()"*)&+," representation.
$%&'()"*)&+,"
and that
Slemma
anduse
$%&'()"*)&+,"
that
the S to the
pro-
this prove pro-that
lemma
and S to the prove pro-that
and S the pro-
1be an $%&'()"*)&+," $%&'()"*)&+," $%&'()"*)&+,"
$%&'()"*)&+," $%&'()"*)&+,"
Za Zb|⇥i p 1 Z⌘ [!]Z1b |⇥i
pk=1 Za[!] |ki
⌘ Zbp [!]
Let |ki g !|ki u(g) . be
Let an g ! u(g) be an
|⇥iZda[!] Zb ⌘
k=1
] d[[!] sek=1
Z a Z d b[!]
a
ss sda |ki
[!]lca
Z Z | |ki e t
] [ b sed]s[!]s
a Z
k=1 b Z a . Zt aLet
s d
|⇥i
sal Z gbe !
[Zk=1
cl
aZ
g | u(g)
n e
⌘
db na d ] k=1atZ p
at
abe n t s
ean
Z b d
e he l
t
[ dn g
|ki ng n
a
Za a Zb|kit
i m n e u
. s
e
Let nh o
t
g c
! g n
u(g) Toi m be u
prove s
an n o
the c result for the general case where the rep-
le representation
irreducible
-to arS h pirreducible
t dn e eh avof
representation
tor group
ptairreducible
ho
representation
S Gttof inn ethe
group
a v m o same
of rmGp
group ecoho-
in ltheoG tssame ih
in athetm coho-
[!]
m
sameese u lcoho- e
sof wihgroup t weo sN u eresentation
w w oN g → V (g) is not irreducible, we first recall
irreducible ",+&)*representation
")('&%$ ",+&) of*d"group
)('"&
a,+%& representation
$)G *in )(the'&%$same
"irreducible of group
,+coho- )*")(G
"representation
& '&%in $ the same Gcoho- in the same coho-
lassmologythat the class
mology representation
that thethat gthe!representation
representation V (g) that belongs
g !the V to (g) !belongs to g !toV (g) belongs
mology class thatclass mology
the representation class gA! mology V (g) g class
representation
belongs V (g)that belongs
to
Bthe representation A g !toV (g) Abelongs
that
B according
A to B
B to lemma AB 5, all the  irreducible projective
lemma it. From 5 we it. know
lemma
From that
5 weall
lemma know
it. the
5 we
From irreps
that
know allinthat
lemma the
the 5 same
irreps
all
we theknow inirreps thethat same
in allthe the same irreps in all thethe same A A B B
it. From
ogycohomology
class have lemma the 5 samewe know
dimension, that all and the irreps
it.
therefore From in the
lemma same 5 we know that  = irreps
 + in the
representations
  = same
 +  of an Abelian group in the same cohomol-
cohomology class have class the have
cohomology same the dimension,
same
class dimension,
=
have and the + therefore
same and therefore
dimension, and therefore A
 B
=  A  + =  B A +  B  =  A +  B
cohomology class have the same dimension, cohomology and therefore
A B
class have the same dimension, and therefore A B  =  + 
is d[!] . Also, note ogy class are equivalent up to a unitary and a phase. This
nsion A B
the of this
dimension
the representation
B dimension of thisthe of is drepresentation
representation
this
dimension [!] . Also, dnote
is this . isAlso, d[!] . note Also, note
ethe |⇥i dimension satisfies of this representation B A
isofthe [!] . representation
d[!]dimension Also, note
of
A
this representation is d . Also, note
thata state
Z Z that |⇥iZa Zb satisfies
state |⇥i satisfies
a Zb state |⇥iZa Zb satisfies FIG. 7: By consuming
Zthat
[!] result implies that
|⇥i, entangled any projective unitary representation
Bya consuming
maximally aentangled state Alice
b
that state |⇥iZ B a Zb + A
satisfies = B +that A state =|⇥iZa Zb satisfies FIG. 7: FIG. 7: By consuming FIG.maximally 7: aBy maximally
entangled
consuming state
a maximally|⇥i,
|⇥i,
stateAlice |⇥i, Alice state |⇥i, Alice
entangled
⇤ FIG. ⇤
7: By consuming the maximally and Bob can FIG.
entangled
prepare
and
7: Bystate
Bob and can
consuming
state
Bobprepare
(ab)
can gaand
, which
prepare
state
→
maximallyV(ab)
is
Bob
(g)
equivalent
state ,can
ofentangled
which
(ab)anFIG.
prepare,
isto
which 7: state
Abelian
state
equivalent
state
Byis group
consuming
(ab)
equivalent
to ,
Alice
state
which
G,to induces
a maximally
isstate
equivalent
a to tensor
entangled
state
prod-
state |⇥i, Alice
u (g) ⌦ u(g)]|⇥i ⇤[u (g)
⇤ Z⌦ a[uZu(g)]|⇥i= |⇥iZZa ZZb .[u
b (g) ⌦ u(g)]|⇥i
⇤
= ZBob |⇥i
(g) ⌦a= (5.13)
|⇥i
u(g)]|⇥i. Za ZbZ⇤a.⌦ = |⇥i
(5.13) (5.13)
Z)aand .toBob can (5.13)
prepare state (ab)
, whichand is equivalent to state (ab)
|Θi
[u (g) ⌦ u(g)]|⇥i Z a Z b , Alice a and
Za Zb = |⇥iZa Zb .
b a ZbZ can
Z b measure
[u (g)
Z(AB|C)
(5.13)
b the
⌦ u(g)]|⇥i total
( up
Z bcharge
(AB|C)
⌦(AB|C)=( |⇥i local
⌦ in
(AB|C) sys-
operations.
) up ( )
tob local. up uctto
⌦ decomposition
(AB|C)
local
operations. (5.13) (operations.
) up to Bob
local of the
can
operations. Hilbert
prepare state space , as
which H ≡
is M ⊗
equivalent N ,to state
0 0 (ab) ⌦ Za Zb ( ) up Zato Z local operations. ⌦(AB|C) ( )M up to local operations.
ns Thisthat means if This
we assume tems
means a the and
that b , and
representation
This if we means assume thereby of
the prepare
sym-representation state σ of sym- , which is equiv- where the subsystem corresponds to g → u(g), an
that if we assume the that if we
representation assume of sym- the representation of sym-
This
system means thatalent !if we to⇤ (g) assume
state Ω(AB|C)
the representation
!(Ψ) uThis ⇤ up means to glocal ofthat sym-operations.
⇤ if weZ assume is The theidea Zbis is irreducible
representation of sym- projective representation in the same equiva-
metry Zon a is
metry g on
system u
system
Za is metry and
gZ! a on on
isu⇤g(g)
⇤ system
system and Zon
(g)
Z b and
a issystem
is on! system
uZ b(g) is and b on system
,metry
then ongsystem
state ! |⇥i that
u(g), Z by
then
a has is g!g
adding !
state
charge u
u(g),|⇥i (g)
systems
zero. Zthen and
Zb statehas Z
metryon
a to system
charge|⇥i the
on zero. Z
Alice’s
system b
has is Z side
charge
a is g and !
zero. Z
u ⇤
b (g)to the
and on system lence Zb is of the representation g → V (g), and N is the
class
g ! u(g), then Za Zstate b |⇥i Za Zb has a charge zero. Z a Z b
owsg !that u(g),tothen Bob’s
Itperform state side,
follows |⇥i
thethat athe
Zto
Zcharge has
perform charge
representations
measurement g the!zero. u(g),
charge ofthen symmetry state
measurement |⇥i GZaon both
has chargesys- zero. corresponding multiplicity |⇥i subsystem. In other words,
It follows 0 0that to
It
perform bfollows
the charge that to perform
measurement the charge Z measurement
b LOCC
⌦(AB|C)
LOCC
0 perform 0 0 0 2. Proof of ⌦(AB|C) 2.( )Proof !of|⇥i ( 2.
LOCC ( ) of ⌦!
Proof (AB|C)
( )
LOCC
! |⇥i
 It follows ˜ (athat b ) to
tems the charge 0It measurement (AB|C)
2 ˜ Q}0 0 {⇧
0 b )on systems a {
0 (a
: a 2 Z0 aQ} and
⇧˜and(aon
bb 0b
, Z
) systems
Alice
: b0
become
2 and
Q} a 0 and
Bob
on non-projective.
follows b
systems
0
, that
Alice a 0 andto
and Therefore,
perform
Bob
b 0
, Alice the and Alice
charge
2. Bob Proof
2. Proof of ⌦
measurement
of ⌦(AB|C)
there exists ( )
)a LOCC ! |⇥i W such
unitary! |⇥i 2. Proof of ⌦ that (AB|C) LOCC
˜{(a ⇧b ) :  and 2 Q}Bob on can systems 0a and 0b˜ ,(aAlice b )the and Bob in a0 Za and ! |⇥i
 0 0 ( )
{⇧  2charge
: can Q} onmeasurements
perform systems
the local locallyacharge and measurebsystems
,local
Alice
measurements and charges Bobon on systems 0 b0 Zb .0
orm the local
can perform the local cancharge perform on
measurements the { ⇧  :
charge
on  2
systems Q}
measurements systems ona and
systems b , Alice and Bob
0 But, 0
b because the |Θi In the following, In
similar the to following,
the approachIn similar
the we to
following,usedthe approach
in the
similar to wethe used
approachin
X the we used in the
0
can (See aFig.7).
b Zb0 perform Z0athe andMore local Zb precisely,
(See
charge
aZ 0 Fig.7).
and btotal
measurements
Alice
0 a0 More precisely,
0
ZMoreb (See
charge
can
performs precisely,
perform
Fig.7). on in systems Alice
More the b is charge
performs
local
Zaprecisely,
Z zero, In
Alice thethe
measurementstotal similar
following,
performs on systems
to the approach LOCC we†used in the iθr (g)
0 a Za and the0 b measurement
Zcharge
b (See Fig.7). in a b corresponding
is equal to 0 to
the Alice
the
sum performs
projectors
0 of the
previous charges section, Ininthe a
we 0 following,
previous
Z prove and similar
section,
(ab) ∀g
LOCC we
!∈toG the
prove
|⇥i, : andW
approach V
(ab) (g)W
In
this the we
to- ! =
used
following,
|⇥i, u(g)
in
(ab)and the⊗
LOCC
similar
this! to- e
to
|⇥i, the |rihr|
approach , we (5.18)
used in the
a
urementZ a and b Z
corresponding
b (See Fig.7). tothethe More precisely,
projectorsa Z
measurement Alice
a and b Zb (See
corresponding performs toFig.7).
the previous More precisely,
projectors a
section, we Alice previous
prove performs (ab) section,
LOCC we
! |⇥i,
prove
and thisLOCC to- r
and this to-
the measurement b0 Zbcorresponding corresponding
, denoted by κ toand theκprojectors in the above figure. previous section, we prove (ab) LOCC LOCC !
previous |⇥i,
(AB|C) and
section, this weto- !
prove |⇥i. (ab)LOCC LOCC
! |⇥i, and this to-
the measurement to
A the theprojectors
B measurement corresponding
gether with proposition 7 imply ⌦ gether
to the with
projectors proposition
(AB|C)
gether 7
( ) (AB|C)
with imply ⌦
! |⇥i. LOCC
proposition 7 (
imply ) ⌦ (AB|C)
( ) ! |⇥i.
1 X ˜ (a0 Za ) X 1 X 0 i(g) 1 X ⇤ gether with proposition 7 imply ⌦(AB|C)((ab) ) LOCC ! |⇥i.
2 (ab) LOCC LOCC
= ˜0 (a0 Za )e⇧i(g) 1X V =0 (g)⌦u
a |G| i(g) ⇧˜ ⇤(a eZa )
(g) : =
 2 VQ a0 (g)⌦u
, (5.14) e (g) :V
i(g)
(g)⌦u
a First,
0
Q ,
X we prove
⇤(5.14)
(g) gether :  2 with
Q First,
(ab), proposition
(5.14)LOCC we ! (ab) prove
|⇥i, 7
First,imply
forgthe ⌦
we special !gether
proveacts |⇥i, ( )for
with
caseirreducibly
LOCC
the !
proposition |⇥i.
special
! |⇥i,on case
7
forM, imply ⌦
the special case
(AB|C)
( ) ! |⇥i.
˜|G| ⇧Za ) =1 ei(g) 
V 0 (g)⌦u |G|⇤ (g)˜ (a: 0 2 Q , 1(5.14) First, we ⇤prove where LOCC
! → u(g)
|⇥i, for therepresentation
gspecial case LOCCand {|ri} is a ba-
⇧ (a
 g2G= |G| e g2G Va0 a(g)⌦u⇤ (g) g2G ⇧: Z2a )Q=, where (5.14) the e i(g) Va0 (g)⌦u
First,
projective where
we prove the: projective
(g)
representation  (ab)
2whereQ ,LOCC g (5.14)
! representation
the
! V projective
|⇥i,
(g) for
of the
group
First,
! V (g)
special
we prove case of
(ab)group
g ! V (g)
!arbitraryof group
|⇥i, for the special case
|G| g2G |G| where the G on projective
the virtual sisG
for thevirtual
representation
systems
on therepresentation,
subsystem ! V (g)
is an ggirreducible
systems
N ,of and group eiθr (g) are
representation, phases.
0 on systemsg2G
a 0
and Z and obtains
0 charge G  ong2G
. the virtual
Similarly, wheresystems the projectiveis an irreducible
representation where ! Vthe (g)is an
of
projective irreducible
group representation,
representation g ! V (g) of group
ms a and Za and This
0 obtainsthe lemma on basically
systemsa
charge A . Similarly, a andmeans Z a andthat obtains
Afor
and thencorre- the charge G
composite
G on A
explain on  the
the
.
howand virtual
Similarly,
virtual sys-
then
this argument systems
explain
systems and Using is
howthen
iscan
an
an be
irreducible
this
thisexplain observation,
argument
extended
irreducible how representation,
can
tothis
representation,
be wesystems
argument can extend
extended can to the above
be irreducible
extended proof
torepresentation,
Gcanisonbasedthe virtual is an
orms
on systems the Bob
on systems local aperforms
a0 tem and
and Za Bob
projective Zwith
and local
obtains
performs
measurement
a andrepresdentation
obtains
projective
charge
charge theon corre- 
systems
measurement
local
gA. →
A
. projective
Similarly,
Similarly,u(g)
a
the
0
and ⊗Zu
general
∗ and then
measurement
and(g),
and case. any
obtains
then
theexplain
The statecorre- how
general
charge
explain proof howis
case.
to.thethis
based
thisthe
Similarly,The
argument
case
generalon
argument
proof
where
the case.following
can the
Thebe
on the
be representation
extended
proof
extended ishow following
to
based
to ong →theV following
(g) is not irre-
sponding to the projectors a A and then explain this argument can be extended to
toBob
Bob the performs the local
projectors
performs the local
with a sponding
definite
projective
projective chargeto the measurement
measurement
is
Bobprojectors
a maximllay
performs lemma
corre-
corre- the entangled
local the general
projective lemma
state.
the general case. The proof is based
case. The
measurement lemma
ducible: proof corre- is based on
Because onthe the following
initial
the following states |λi 0 and |λi 0 both
sponding to the projectors the general case. The proof aa is based on b bthe following
sponding to˜ (bthe projectors
Zb )This 1 X 0 i(g) X spondingthe to the projectors lemma
g → Lemma ⊗ 9be Let g ! u(g) be g an! ifirreducible projective
1 X ⇧ =⇤ lemma follows V1b⇤0 (g)from ⇤fact2 Q that 9 lemma u(g) have charge zero, the
u(g)charge measurement on systems
0
(beZb )
˜u(g) ⌦e u(g) i(g) : Lemma , (5.15) Let g ! u(g) an Lemma irreducible 9 Let lemma
projective be an irreducible projective
= e i(g)
 ∗ X
V (g)
|G| ⌦ ⇧ : =
 2 Q , (5.15) V 0 (g) ⌦ u(g) :  2 representation Q , (5.15) 0 of 0 an Abelian group G. Then, forG.anyThen, for any
0
˜0 Zb ) =1 u1 X (g) b 0
is a (linear)

⇤ unitary
|G| representation b
representation
X of Lemma
group of an G, 9 Let
and
Abelian g ! a
group u(g)
b projects
representation
G. be an
Then, irreducible
them
offor an to
any Abelian projective
the subspace
group with charge κ, then
˜|G| ⇧ (b Z b ) i(g)
ei(g) V g2G
b0 (g) ⌦ u(g) : 0Zb2 Q ,1 (5.15) i(g) Lemma 9 Let g ! u(g)
⇤ 1-dimensional representation be an irreducible Lemma
g ! e9i(g) projective
Let g any ! u(g) G,beofangroup irreducible projective
⇧ (b g2G= |G|
since ethe group Vb⇤0 (g) is ⌦Abelian
u(g) g2G
⇧˜: (bany 2)Q= such , 1-dimensional
(5.15) e
representation representation
V 0 (g)1
representation ⌦
can P u(g) be of:i(g)
ang 2 Abelian
1-dimensional
Q
systems
! ,P e i(g)group representation
(5.15) ab ofshould G.
group Then,
be
G, in forof
charge g anygroup
!−κ ei(g)
sector. Then, G,if Alice

|G| g2G 0 
|G| P representation b eof an [uAbelian
1 ↵ (g)⌦u group
⇤
(g)]
i(g) isG.
representation
either
i(g) Then, zero,
⇤ forof
or a an rank Abelian
one group G. Then, for any
1-dimensional ⇤ representatione g ![ue
on systems g2G
decomposed b on andsystems Ztob , 1-dimensional
andb0obtains Zb ,charge irreps 1 g2G
Bas . Finally,
g e charge
i(g)
[uP(g)⌦u
1-dimensional
|G| g
↵ (g)] is either
representation and |G| Bob zero,
g
↵
or!
gperform
(g)⌦u
a 1-dimensional
rank oneof
e projective
↵i(g) group
(g)]
of↵|⇥()i
group
G, zero, or a rank
is either
measurements
G, g !onei(g)
one
multiplic-
ms b0 and they Zb , and add obtains
these charge
charges B . and
together, Finally, and
and
obtain
obtains
|G|
the total 1 P↵Bprojector
e
. i(g)
Finally,
[u |⇥()ih⇥()|.
(g)⌦u ⇤
(g)] is Furthermore,
either P
zero, or a rank representation
is
one a maxi- of group G,
on systems
these charges b0together,
and Z they , and add
obtains these charge charges  together,
. projector
Finally, and 1
|⇥()ih⇥()|.
obtain
|G| g the i(g)total
g emally entangled [u↵↵(g)⌦u
Furthermore, ityprojector (g)] is |⇥()ih⇥()|.
⇤↵ |⇥()i
subsystems either is a 1N maxi-
zero, ofg ethe
ori(g)Furthermore,
a virtual
rank onesystems
[u↵ (g)⌦u ⇤|⇥()i
↵ (g)] isaeither
is
and a maxi-
b,zero,
they or will
a rank one
on systems charge b0 and  =Zb ,Aband +and . obtain
∗Bobtains the
charge
+on
total BB. Finally,
systems
iκ(g) b0 totaland,entangled
Zb ,∀g |G|
and
projector obtains charge Bmally
|⇥()ih⇥()|. . ↵state.
Finally,entangled|G|state.
Furthermore, |⇥()i isthe a maxi-
they add these [u(g) charges ⊗ u
charge (g)]|Θ(κ)i
together, = Aand = Be.
obtain |Θ(κ)i
the mally ∈ Gstate. (5.16) find the M
= 
charge
+
theyA addBthese 

. Then,charges
=  +
using the
 .
together,
fact
Then,has that
using andstate
the they |⇥i
obtain
fact addZ the
a bthese
Z
that has total
charge
statecharges |⇥iZamally
projector
zeromallytogether,
hasentangled This |⇥()ih⇥()|.
entangled
charge and lemma
zero obtain basically
state.This the lemma total basically means that for the composite sys- isofa maxi-
Furthermore, total
means charge
that |⇥()i
projector for of
the is a
|⇥()ih⇥()|. irreducible
maxi-
composite sys- subsystems
Furthermore, |⇥()i
using
charge  = the fact
with thatA
Arespect state
+ B . towith B |⇥i theaiκ(g)
Z representation
Z charge zero
g != u  ⇤ This
(g) ⌦ lemma
u(g)
Z
(See bbasically means state.
that systems
for the a and
composite !b. u(g)
g mally But,
sys- ⌦ from
entangled u⇤ (g),state. lemma 9 we know that any
charge hascharge A+ Bg. ! u⇤ (g) ⌦tem ofwith (See represdentation any state
b
ectThen,toThen,theusing using where
representation the fact g→ that
g! e respect
u⇤ (g)
state isadding
⌦a
|⇥i tou(g) the
1-dimensional
a Zb(See
representation
tem representation
zerowith This
represdentation lemma u(g) G.
basically
g ! tem
u(g)means ⌦with u ⇤ represdentation
that
(g), for
any thestate composite g !sys- u(g) ⌦ u⇤ (g), any state
Eq.(5.13)), the fact we that find
Eq.(5.13)),statethat |⇥i Z systems
has
b Then,
⇤ thatusing charge Z azero
the and Z
fact that bThisZ in lemma
state with |⇥iZ a definite
basically state
has charge charge
means of
agzero
ab
is a which
maximllay
that forThis has
thelemma a definite
entangled
⇤composite basically charge
state.
sys- means should
that state.for be theacomposite
maxi- sys-
Zagwe gZ→ find adding systems a and Zba Zin
aZ with definite charge is aany maximllay entangled
), with
with werespect respect
find state that Then,
to|⇥i
toadding theZthe
theZarepresentation does fact
representation
systems notthat change gand
!
! the
u
u(g)
u
Z
⇤ (g)
btotal
(g) in⌦is u(g)
anwith
⌦charge.
u(g) irreducible
(See
(See aHowever,
definite temrepresenta- with
charge represdentation
This is! a lemmab
maximllay
⇤ follows !from
entangled u(g)the ⌦ ufact
state. ⇤(g),that gstate
! u(g) ⌦ LOCC
state |⇥iZa Zsystems with
does not respect to the tem
representation with represdentation
g u (g) ⌦
mally u(g) g !
(See
entangled u(g) tem ⌦ uwith
state. (g), any
represdentation
This state
proves g !
thatgσ! u(g) (ab)
u(g) ⌦
−−−⌦ u ⇤
(g),
−→ |Θi, any state
Zachange theintotal charge. However,
b
Eq.(5.13)), we find
tion, that adding and Z This lemma ! follows ⌦from the fact that
find together with the Schur’s lemma b Zb implies
athat
does by notwe change the total Zcharge. b However, withfollows au ⇤ each
definite
Z a Zb
Eq.(5.13)), adding systems
that by adding
adding a tosystemsthe
systems
Alice’s
Eq.(5.13)), Z
Z
side
andthe
to
and
weZ This in lemma
find
bAlice’s
to
that the
with
side adding
and definite
Z (g)
systemsfrom
to istheacharge
charge the
(linear)
Z
is
fact
is
and
a maximllay
⇤ aunitary that
maximllay
Z in with
entangled
grepresentation
u(g)
entangled
a definite
state.
of group
state.
charge G, is and
a maximllay entangled state.
gstate state |⇥i
systems ZZaa Zto does
thenot not
Alice’s change side the total charge. a
a However, ⇤ b aandu (g) is
completes
b a (linear) the unitaryproof
g ! of representation
theorem. of group G, and
|⇥iZBob’s a Zb
state
b side,
does |Θ(κ)i
the
Bob’sshould
change theand
representations
side, total
the
Zb charge.
be to
stateaof the
maximally
symmetry
representations|⇥i However,
Z Z
u (g)
does
Gentangled
of not
bothThisstate.
isona (linear)
symmetrychange This the
⇤ G
lemma
unitary
since
lemma total
on theIt
both
follows
representation
group
follows
charge. is
since
from
fromAbelian
However, the
thegroup
of
thegroup
factsuch
factisG,
any that
that
This
and
Abelian
representation
g !
lemma any
u(g)
u(g)
follows
such
⌦ can be
⌦representation
from the fact can that
be g ! u(g) ⌦
de, by adding systems Zof a to the Alice’s side and Zbsince to → the
by the representations
sides become symmetry
(linear) non-projective G on both representation: u (g)
On uisisis a (linear) anyunitary representation ofbe group G, and
a b ∗
the group Abelian
decomposed such representation
to 1-dimensional ⇤ can
irreps as
adding systemsfollows Za that to
sides the inbecome
the
Alice’s decomposition
side
(linear)
by adding andnon-projective
Zbsystems tog the u(g) uto⇤⊗
Z representation:(g)the (g) each
a (linear)
Alice’s On unitary
side and Zb representation
decomposed to the tou1-dimensional (g)ofis groupa (linear) G,irreps
and
unitary as representation of group G, and
ome Bob’s(linear) side,
Alice’s theside
non-projective representations
the representation representation: of symmetry is g ! OnVG a0 (g) on ⌦both
decomposed u⇤ (g),a and
iκ(g) since
to the
1-dimensional group is Abelian
irreps as any such representation can be
Bob’s side, the 1-dimensional
representations Alice’s representation
sideof symmetry
the⇤ representation
Bob’s Gg on →both eisrepresentations
g !can Vsince
a0 (g) show ⌦ uof
the ⇤ up at
(g),
group
[u(g) and⌦ ⇤
is uAbelian G onany
(g)]|⇥()i such
= ei(g)representation
|⇥()i the=,group 8gcan Gbe (5.16)
2 Abelian
de sidesthe representation
on Bob’s
become (linear) side is gthe ! Vrepresentation
non-projectivea0 (g) ⌦ u (g), is side,
and
representation: g ! the Vb⇤0 (g) On⌦ u(g). ⇤decomposed
symmetry
toei(g)
1-dimensional
both
[u(g) ⌦irreps since
u⇤ (g)]|⇥()i
as (5.16) ei(g)is|⇥()i , any 8g 2 such G representation
(5.16) can be
sides become (linear) most once. non-projective
on Bob’s side
⇤ representation:
the sides representation
become ⇤ On is⌦gu
[u(g)
(linear) ! Vb⇤0 (g) ⌦=to
(g)]|⇥()i
decomposed
non-projective u(g). |⇥()i
1-dimensional
representation: i(g) VI. , SPT-ENTANGLEMENT
On8g
irreps2 G as
decomposed to 1-dimensional IN ARBITRARY
irreps as
sideAlice’s the Therefore,
representation
side the representation because is g the ! is Vlocal
g (g)
! representation
⌦
V u(g).
0 (g) ⌦ u (g), of
and symme- where g ! e is a 1-dimensional representation of G.
u⇤local i(g)
b! 0 a ⇤
Therefore, because 0 (g)the ⌦⇤side representation of⌦gsymme- ⇤ ei(g) !
Alice’s side try the representation
onlocal From
Alice this
and lemma
Bob’s is gsystems we VAlice’s
afind are Vthat
(g),
ifthe the and
representation
representation ei(g) [u(g) isis auu !⇤g (g)]|⇥()i
→V a0 (g) ⌦where
u= = (g),
ei(g)
g!
gand e
|⇥()i
u(g) is ,,[u(g)
is a 1-dimensional
anof8g DIMENSION
2G (5.16) representation of G.
e, because the the representation of gsymme- ⇤bnow 0 (g) where ⌦Abelian, g !they Then, the
1-dimensional fact that representation irreducible
G. representa-
⇤
on Bob’s
on Bob’s can sideV
side
the
representation
representationtry on Alice is
isand
g !
!Bob’sV (g) systems
⌦
u(g).
u(g). are now [u(g) Abelian, ⌦ (g)]|⇥()i
they ⇤ Then, the |⇥()i
fact that g 8g ! ⌦2u(g) uG (5.16)
(g)]|⇥()i
is an = ei(g) |⇥()i
irreducible representa- , 8g 2 G (5.16)
liceTherefore,
and Bob’sbecause (g) is the
measure
systems an areirreducible
the charges
now
local Abelian, locally
representation on Bob’s
projective
they and
b 0
ofaddside Then,
symme- the
representation,
these representation
the local
fact where thatthese gthen
tion,
g! is
!ei(g) g
i(g)
u(g) !
together V
is an 0 (g)
ba with ⌦
irreducible
1-dimensional u(g).
the Schur’s representa- lemma implies
representation that each
oflemma
G.
Therefore, because the local can measure
representation the charges
Therefore, of symme- locally
because and
thewhere add
local !
grepresentation local
e Schur’s tion,
is a 1-dimensional
of u(g) symme- together with the
representation
where Schur’s
g !representa-
e i(g) ofisG. implies that
a 1-dimensional representation each
sure trythe on charges
try on Alice and Bob’s
Alice and locallyBob’sX and systems
systems
add
are
these
are
now
now local Abelian,
Abelian,
tion, they
they
together Then,
Then,
with the
the
thefact
fact
that glemma
that g !
! In u(g)
implies
this is an
is ansection that
irreducible
irreducible
each
we study
representa- some general properties of of G.
(ab) try
(KC ) on Alice and Bob’s systems are now Abelian, they Then, the fact that g ! u(g) is an irreducible representa-
can measure theσ charges
can measure the = locally
charges pκ |κihκ|
locally and
andcan
add
addmeasure ⊗these |Θ(−κ)ihΘ(−κ)|
these local
local
tion, together
tion,ab . (5.17)
together
with the Schur’s lemma implies
SPT-Entanglement, which that hold each in arbitrary dimension.
the charges locally andwith addthe these Schur’s locallemma tion,implies together that with each the Schur’s lemma implies that each
κ In particular, we study how SPT-Entanglement changes
under the effect of low-depth symmetric circuits.
This is true because if register KC has value κ then this
means that the charge measurement on a0 b0 has projected
them to the charge sector with charge κ. But, because
states |λiaa0 and |λib0 b both have charge zero, the initial A. Monotonicity of SPT-Entanglement with the
total charge of systems aba0 b0 is zero. This means that size of regions A and B
if the systems a0 b0 are projected to state with charge κ
then the systems ab should be projected to state with Our first result states that as we make regions A and
the total charge −κ. Then, by lemma 9 we know that B larger, SPT-entanglement either remains constant or
there is a unique state of ab with charge −κ, namely state increases. More formally,
 19

Proposition 10 Let A0 and B 0 be a pair of non- value of this register, i.e. they implement the following
overlapping regions that contain A and B respectively, transformation
that is A ⊆ A0 and B ⊆ B 0 . Let C be a region sur-
rounded by A, B and the boundaries of the system, and |κC 0 i(KC 0 ) −→ |κC 0 + κ∆A + κ∆B (mod 2π)i(KC ) . (6.5)
C 0 ≡ C \ (A0 ∪ B 0 ) be the subregion of C surrounded
by A0 , B 0 and the boundaries of the system (See Fig.8). Finally, by tracing over the sites in regions ∆A and
0 0 0
Then, ∆B they can transform state Ω(A B |C ) (ρ) to state
Ω(AB|C) (ρ). Note that to find the total charge in region
0 0
|C 0 )
Ω( A B
LOCC C and update the classical register, Alice and Bob need
(ρ) −−−−→ Ω(AB|C) (ρ), (6.1)
to have classical communication with each other.
Therefore, for any measure of entanglement E, it holds
0 0 0
that E(Ω(AB|C) (ρ)) ≤ E(Ω(A B |C ) (ρ)) .
B. Effect of low-depth symmetric circuits on
If we choose negativity as a measure of entanglement then SPT-Entanglement
this implies

iTA

i TA
Next, we consider the effect of low-depth symmetric
X
h
X
h


tr (Π(C)

(C 0 )
. circuits on SPT-Entanglement, and show that under such

AB κ ρ)
≤
trA0 B 0 (Πκ ρ)
transformations SPT-Entanglement cannot change dras-
κ 1 κ 1
(6.2) tically.
Note that here we are looking at the bipartite entangle- Let A and B be two non-overlapping regions of the
system and C be a region surrounded by A, B and the
boundaries of the system. In the following, to simplify
the presentation, we assume these three regions cover the
entire system, and there is no site outside these regions.
For any region X let XR be the set of sites in the ball of
radius R from X, that is the set of sites whose distance
from X is less than or equal to R. Similarly, for any re-
gion X let X−R be a subset of X whose distance from the
FIG. 8: Region A0 is partitioned to region A and region ∆A.
complement of X is larger than R. Note that region C−R
Similarly, region B 0 is partitioned to region B and region ∆B.
Region C 0 ≡ C \ (A0 ∪ B 0 ) is the subregion of C surrounded is surrounded by regions AR and BR , and the boundaries
by A0 , B 0 and the boundaries of the system. of the system, and similarly region CR is surrounded by
regions A−R and B−R , and the boundaries of the sys-
tem. Therefore the sets {A, B, C}, {AR , BR , C−R } and
ment of state {A−R , B−R , CR } define three different ways of partition-
X ing the sites of systems to three regions.
0 0
|C 0 ) 0
Ω(A B (ρ) = |κihκ|(KC 0 ) ⊗ trA0 B 0 (Π(C
κ
)
ρ) , (6.3)
κ
Theorem 11 Let V be a symmetric circuit with range R
which is bounded by R < dist(A, B)/2. Then, the follow-
where the systems KC 0 , A0 and B 0 are partitioned as ing transformations can be implemented via LOCC,
A0 |KC 0 B 0 (or equivalently A0 KC 0 |B 0 ). In other words, all
the sites in region A0 are given to Alice, and all the sites Ω(AR BR |C−R ) (ρ) →
− Ω(AB|C) (V ρV † ) →
− Ω(A−R B−R |CR ) (ρ) .
in region B 0 are given to Bob, and the classical register (6.6)
KC 0 , which keeps the information about the charge in
region C 0 , is given to either Alice or Bob. That is there exist LOCC transformations which trans-
The proof of proposition 10 is simple: Let ∆A = A0 \A form Ω(AR BR |C−R ) (ρ) to Ω(AB|C) (V ρV † ), and the latter
and ∆B = B 0 \ B be, respectively, the part of region A state to Ω(A−R B−R |CR ) (ρ). Therefore, for any measure of
which is not in A0 , and the part of region B which is not entanglement E,
in B 0 (See Fig. 8). Alice and Bob, who are given all the E(Ω(A−R B−R |CR ) (ρ)) ≤ E(Ω(AB|C) (V ρV † ))
sites in regions A0 and B 0 respectively, can measure the
charge of the sites in regions ∆A and ∆B. Then, because ≤ E(Ω(AR BR |C−R ) (ρ)) . (6.7)
the charge is Abelian, the total charge in region C is
In words, this inequality means that the SPT-
κC = κC 0 + κ∆A + κ∆B (mod 2π), (6.4) Entanglement of state V ρV † for regions A and B is lower
bounded by the SPT-Entanglement of ρ between A−R
where κC 0 is the charge in regions C 0 , κ∆A is the charge in and B−R , and upper bounded by the SPT-Entanglement
region ∆A, and κ∆B is the charge in region ∆B. There- of ρ between AR and BR .
fore, adding the charges κ∆A and κ∆B to the charge in In Sec.(III B) we sketched a proof of this theorem.
region C 0 , whose value is written in the register KC 0 , they Here, we present another proof, which is based on a
find the total charge in region C, and they update the slightly different point of view. (III B).

Proof. The key point in the argument is that because
V is a low-depth circuit with range R, having access to
all the sites in the extended regions AR and BR , Al-
ice and Bob can apply all the unitaries in this circuit
which are in the light cones of region A and region B
(See Fig.(1)). Call these local symmetric unitaries which
act non-trivially on AR and BR , VAR and VBR , respec-
tively. Then, for any operator X,



TrAB V XV † = TrAB [VAR ⊗ VBR ] X [VA†R ⊗ VB†R ] ,
(6.8)
where the partial traces is over all sites in the system ex-
cept those which are in A and B. Note that this equation
is simply a consequence of the fact that V is low-depth.
In particular, applying this equation for X = ρ we find



TrAB V ρV † = TrAB [VAR ⊗ VBR ] ρ [VA†R ⊗ VB†R ] .
(6.9)
Next, using the fact that both unitaries V and VAR ⊗VBR
are symmetric, and hence preserve the total charge, to-
gether wit the fact that the reduced state of regions A and
B is the same for states V ρV † and (VAR ⊗ VBR )ρ(VA†R ⊗
VB†R ), we can easily show that the charge in region C is
also the same for these two states (Note that the charge
in region C can be thought as the total charge in the
system minus the charge in regions A and B). In other
words, applying Eq.(6.8) for X = Πκ ρ, where Πκ is the
projector to the subspace with charge κ in the system,
and using [Πκ , V ] = [Πκ , VAR ⊗ VBR ] = 0, we can easily
show that

that the effect of symmetric low-depth circuits on SPT-
† † †

TrAB Π(C)κ V ρV = TrAB Π(C)
κ [VAR ⊗VBR ] ρ [VAR ⊗VBR ] ,
(6.10)
(C)
where Πκ is the projector with charge κ in region C.
Using this observation we can easily find a LOCC pro-
tocol which transforms state
X
Ω(AR BR |C−R ) (ρ) = |κihκ|K ⊗ TrAR BR (ρΠ(C −R )
),
κ
κ theorem implies that the SPT-Entanglement should re-
(6.11)
to state
X stant throughout a phase.
Ω(AB|C) (V ρV † ) = |κihκ|K ⊗ TrAB (Π(C) †
κ V ρV ) .
κ showed in proposition 10 by making regions A and B
(6.12)
The protocol includes the following steps: (i) Alice and
Bob apply the local symmetric unitaries VAR and VBR
defined above on the sites in regions AR and BR . (ii)
They measure the charges in regions ∆A ≡ AR \ A and
∆B ≡ BR \ B. (iii) They add the charges obtained in
these regions to the charge in region C−R which is written
in register K, to find the total charge in region C for state
ρ, and write down this charge in register K. (iv) They
discard all the sites in regions ∆A, and ∆B.
One can easily show that this protocol implements the
following transformation
LOCC
Ω(AR BR |C−R ) (ρ) −−−−→ Ω(AB|C) (V ρV † ) . (6.13)
20
Next, to prove the rest of the theorem, we use the fact
that if V is a symmetric circuit with range R, then V † is
also a symmetric circuit with range R. Then, applying
the above result with V † instead of V , and V ρV † instead
of ρ, we can easily show that
LOCC
Ω(AB|C) (V ρV † ) −−−−→ Ω(A−R B−R |CR ) (ρ) . (6.14)
This completes the proof.
1. SPT-Entanglement vanishes in the trivial phase
Using this result, we can easily show that SPT-
Entanglement should vanish for all states in the trivial
phase.
Corollary 12 Suppose there exists a symmetric circuit
with range R, which transforms state ρ to a product state.
Then, the SPT-Entanglement of state ρ between any two
regions A and B with distance more than 2R is zero,
i.e. for any measure of entanglement
E, it holds that
dist(A, B) > 2R implies E Ω(AB|C) (ρ) = 0.
2. SPT-Entanglement is universal in all phases
Corollary 12 implies that in the trivial phase SPT-
Entanglement remains constant throughout the phase.
This result was a corollary of theorem 11 which shows
Entanglement can be bounded by looking to the small
deformations of the boundaries of the regions A and B.
Next, we consider the consequences of theorem 11
for SPT-Entanglement in a general, non-trivial SPT
phase. In particular, we show that, assuming the SPT-
Entanglement remains unchanged under small deforma-
tions of the boundaries of regions A and B, then this
main constant under the effect of symmetric low-depth
circuits. That is the SPT-Entanglement should be con-
To explain the assumption, first recall that as we
larger SPT-Entanglement cannot decrease, i.e.
0 0
|C 0 )
A ⊆ A 0 , B ⊆ B 0 ⇒ Ω( A B
LOCC
(ρ) −−−−→ Ω(AB|C) (ρ) .
(6.15)
The above relation holds for arbitrary state ρ and arbi-
trary regions A and B. For a general state, the SPT-
Entanglement between the larger regions A0 and B 0 is
larger than the SPT-Entanglement between the smaller
regions A and B, and hence this transformation is not re-
versible via LOCC. However, assuming that the system
is sufficiently homogenous and regions A and B are suffi-
ciently large and far from each other compared to the cor-
relation length, one expects that the SPT-Entanglement
should saturate, and therefore a small increase in the

sizes of regions A and B does not increase it anymore.
In other words, it seems natural to assume that in this
limit the inverse transformation should also be possible
0 0 0
via LOCC, i.e Ω(AB|C) (ρ) −−−−→ Ω(A B |C ) (ρ). Note
LOCC
that, as we saw in the previous section, our results based
on MPS representation of SPT phases proves the validity
of this assumption in the case of 1-dimensional systems.
However, the validity of this assumption in the higher
dimensions is still an open problem.
In the following to simplify the presentation we assume
the regions A, B and C include all sites in the system.
For any region X let X+R be the set of sites in the
ball of radius R from X, that is the set of sites whose
distance from X is less than or equal to R. Similarly, for
any region X let X−R be a subset of X whose distance
from the complement of X is larger than R.
Then, using theorem 11 we can easily show that
Corollary 13 Suppose state ρ satisfies the condition
LOCC
Ω(A−R B−R |C+R ) (ρ) −−−−→ Ω(A+R B+R |C−R ) (ρ). That is
the SPT-Entanglement for smaller regions A−R and
B−R is the same as the SPT-Entanglement for the
larger regions A+R and B+R . Then, for any sym-
metric circuit V with range R, the SPT-Entanglement
of ρ and V ρV † between A and B relative to C are
LOCC
equal, i.e. Ω(AB|C) (ρ) ←−−−→ Ω(AB|C) (V ρV † ). There-
fore, for any measure of entanglement E(Ω(AB|C) (ρ)) =
E(Ω(AB|C) (V ρV † )).
VII. CONCLUSION
The fact that symmetry-protected topological orders
should admit an entanglement-based order parameter
measuring the nonlocal entanglement between the edge
degrees of freedom, seems natural. In this work, we pro-
vided an explicit example of such order parameters. Re-
markably, this order parameter turns out to be closely
related to the string order parameters, via Fourier trans-
form.
Another appealing aspect of this work is to show how
the resource theoretic point of view to entanglement
1
Xie Chen, Zheng-Cheng Gu, and Xiao-Gang Wen, Phys.
Rev. B, 83, 035107 (2011), arXiv:quant-ph/1008.3745.
2 7
Xie Chen, Zheng-Cheng Gu, and Xiao-Gang Wen, Phys.
Rev. B, 84, 235128 (2011), arXiv:quant-ph/1103.3323.
3 8
Norbert Schuch, David Perez-Garcia, and Ignacio
Cirac, Phys. Rev. B, 84, 165139 (2011), arXiv:quant-
ph/1010.3732.
4
Frank Pollmann, Erez Berg, Ari M. Turner, and Masaki
Oshikawa, Phys. Rev. B, 85, 075125 (2012), arXiv:quant-
ph/0909.4059.
5
T. Kennedy and H. Tasaki, Phys. Rev. B, 45, 304307
(1992); Comm. Math. Phys., 147, 431 (1992).
6
I. Affleck, T. Kennedy, E.H. Lieb, and H. Tasaki, Comm.
21
could be useful in the context of many-body systems. If
one tries to formulate the notion of SPT-Entanglement
in terms of a particular measure of entanglement, such as
negativity, then understanding and proving the proper-
ties of SPT-Entanglement will be much harder. Further-
more, using the resource theoretic point of view to entan-
glement, enabled us to clearly see how the universality of
the SPT-entanglement relies on the defining properties
of measures of entanglement, namely their monotonic-
ity under classical communication and local operation.
Quantum resource theories have recently attracted a lot
of attention in the quantum information community, and
it is interesting to see if they find other applications in
other fields of physics, such as many-body systems.
Another feature of the SPT-Entanglement is that it
uses measures of bipartite entanglement to capture tri-
partite correlations in the system, that is correlations
between local degrees of freedom in regions A and B,
and a classical degree of freedom in region C, namely the
total charge in this region. It is interesting to see if this
sort of quantification of tripartite correlations via bipar-
tite entanglement measures have any other applications
in many-body systems (See46,47 for a different approach
to tripartite correlations).
Finally, we noted that the identity gate in MBQC can
be interpreted as the charge measurement for the group
Z2 × Z2 . This simple observation may, to some extent,
demystify the robustness of the computational power of
1-dimensional cluster state under symmetric perturba-
tion.
VIII. ACKNOWLEDGMENT
I would like to thank Stephen Bartlett and Paolo Za-
nardi for reading the manuscript and providing very use-
ful comments. Also, I acknowledge helpful discussions
with John Preskill, Lorenzo Venuti, Benny Yoshida, Isaac
Kim, Spyridon Michalakis, Steve Flammia, and Jacob
Bridgeman. This research was supported by ARO MURI
grant W911NF-11-1-0268.
Math. Phys., 115, 477 (1988); Phys. Rev. Lett., 59, 799
(1987).
H. Nonne, P. Lecheminant, S. Capponi, G. Roux, and E.
Boulat, Phys. Rev. B. 2010 Jan 22;81(2): 020408.
K. Kobayashi, M. Okumura, Y. Ota, S. Yamada, M.
Machida, Phys. Rev. Lett. 2012 Dec 5;109(23): 235302.
9
I. Cohen, A. Retzker , Phys. Rev. Lett. 2014 Jan 31;112(4):
040503.
10
M. Endres, et al., Science 334.6053 (2011): 200-203.
11
D. V. Else, S. D. Bartlett, and A. C. Doherty, New J. Phys.
14, 113016 (2012), arXiv:quant-ph/1207.48053.
12
D. V. Else, I. Schwarz, S. D. Bartlett, and A. C. Do-
herty, Phys. Rev. Lett., 108, 240505 (2012), arXiv:quant-
 22

ph/1201.4877. 020506 (2009).

13 43
B. Yoshida, Phys. Rev. B 91, 245131 (2015). R. Raussendorf, D. E. Browne, and H. J. Briegel, Phys.
14
J. Miller, and A. Miyake, Phys. Rev. Let 114.12: 120506 Rev. A 68.2, 022312 (2003).
44
(2015). S. O. Skrvseth, S. D. Bartlett, Phys. Rev. A 80, 022316
15
H. P. Nautrup, and T.C. Wei, Physical Review A 92.5, (2009).
45
052309 (2015): . J. K. Pachos and M. B. Plenio, Phys. Rev. Let 93.5, 056402
16
R. Raussendorf, D. Wang, A. Prakash, T.C. Wei, D. (2004).
46
Stephen, arXiv:quant-ph:1609.07549 (2016). I. H. Kim, Phys. Rev B 89.23 (2014): 235120.
17 47
A. Hamma, R. Ionicioiu, and P. Zanardi, Phys. Rev. A 71, L. Fiedler, Leander, P. Naaijkens, and T. J. Osborne,
022315 (2005). arXiv:quant-ph:1608.02618 (2016).
18 48
A. Kitaev, and J. Preskill. Phys. Rev. Let. 96, 110404 In the phase protected by group G, there are states which
(2006). are protected by larger groups, such as G × G. Here, the
19
M. Levin, and X.-G. Wen, Phys. Rev. Let. 96, 110405 on-site representation of G × G is defined by grouping two
(2006). neighbor sites in the original lattice. Then, if we know the
20
Xie Chen, Zheng-Cheng Gu, and Xiao-Gang Wen, Phys. charge corresponding to group G for all single sites in re-
Rev. B, 82, 155138 (2010), arXiv:1004.3835. gion C, we can also find the charge corresponding to group
21
M. Horodecki, and J. Oppenheim, Int. J. of Mod. Phys. B, G × G. But SPT-Entanglement for the phase protected by
27(01n03) (2013). group G × G could be larger. It follows that having this
22
R. Horodecki, P. Horodecki, M. Horodecki, K. Horodecki, extra information, in some cases we can extract more en-
Rev. Mod. Phys. 8, 865-942 (2009), arXiv:quant- tanglement, and the quantity defined in this way will not
ph/0702225. be universal.
23
G. Vidal, R. F. Werner, Phys. Rev. A 65, 032314 (2002),
arXiv:0102117.
24
L. Amico L, R. Fazio, A. Osterloh, V. Vedral, Rev. of Mod.
Phys., 80 (2):517 (2008).
25
M. Popp, F. Verstraete, M. A. Martin-Delgado, & J. I.
Cirac, Phys. Rev. A, 71(4), 042306 (2005); F. Verstraete,
M. Popp, & J. I. Cirac, Phys. Rev. Let., 92(2), 027901
(2004); L. C. Venuti and M. Roncaglia, Phys. Rev. Lett.,
94, 207207 (2005), arXiv:cond-mat/0503021.
26
S. T. Flammia, A. Hamma, A., T. L. Hughes, X. G. Wen,
Phys. Rev. Lett. 103.26, 261601 (2009).
27
A. Hamma, W. Zhang, S. Haas, DA Lidar, Phys. Rev. B
77.16:155111 (2008).
28
A. Hamma, R. Ionicioiu, and P. Zanardi, Phys. Rev. A 71,
022315 (2005).
29
I. Marvian, under preparation.
30
Y. Huang, X. Chen, arXiv:quant-ph/1401.3820.
31
D. V. Else, S. D. Bartlett, and A. C. Doherty, Phys. Rev.
B 88, 085114 (2013), arXiv:quant-ph/1304.0783.
32
W. K. Wootters, Phys. Rev. Let. 80.10, 2245 (1998).
33
D. Perez-Garcia, M. M. Wolf, M. Sanz, F. Verstraete,
and J. I. Cirac, Phys. Rev. Lett., 100, 167202 (2008),
arXiv:quant-ph/0802.0447.
34
Marcel den Nijs, and Koos Rommelse, Phys. Rev. B, 40,
4709 (1989).
35
Frank Pollmann and Ari M. Turner, Phys. Rev. B, 86,
125441 (2012), arXiv:quant-ph/1204.0704.
36
K. Duivenvoorden, T. Quella, Phys. Rev. B, 88, 125115
(2013), arXiv:1304.7234.
37
Jutho Haegeman, David Perez-Garcia, Ignacio Cirac, and
Norbert Schuch, Phys. Rev. Lett. 109, 050402 (2012) ,
arXiv:quant-ph/1201.4174v2.
38
F. Verstraete, J. I. Cirac, J. I. Latorre, E. Rico, and M.
M. Wolf, Phys. Rev. Lett., 94, 140601 (2005), arXiv:quant-
ph/0410227.
39
D. Perez-Garcia, F. Verstraete, M. M. Wolf, and J. I.
Cirac, Quant. Inf. Comput., 7, 401 (2007), arXiv:quant-
ph/0608197.
40
C. Y. Huang, X. Chen, & F. L. Lin, Phys.
Rev. B, 88(20), 205124, arXiv:1303.4190, arXiv:quant-
ph/1303.4190 (2013).
41
S. Singh, & G. Vidal, Phys. Rev. B, 88(12), 121108 (2013).
42
A. C. Doherty, S. D. Bartlett, Phys. Rev. Lett. 103.2,
 23

Appendix A: Projective irreducible representations of Abelian groups (Proof of lemma 5)

In the following we repeat the statement of lemma 5 and prove it:

Lemma Let g → uβ (g) and g → uγ (g) be two finite-dimensional irreducible projective representations of an Abelian
group G whose 2-cocyles belong to the same cohomology class. Then, their dimensions is the same. Furthermore there
exists a unitary W , and a phase eir(g) such that

∀g ∈ G : uβ (g) = eir(g) W uγ (g)W † . (A1)

Proof. Let ω(g, h) be the 2-cocycle of the representation g → uβ (g), i.e.

uβ (g)uβ (h) = ω(g, h)uβ (gh) . (A2)

Since the 2-cocycle of the representations g → uγ (g) and g → uβ (g) belong to the same cohomology class, we know
that there exists a phase eis(g) , such that the 2-cocycle for the representation g → eis(g) uγ (g), is also ω(g, h), that is

eis(g) uγ (g)eis(h) uγ (h) = ω(g, h)eis(gh) uγ (gh) . (A3)

Let u∗β (g) be the complex conjugate of uβ (g). Then, we can easily see that the representation g → u∗β (g) ⊗ eis(g) uγ (g)
is a (non-projective) unitary representation of G, that is its 2-cocycle is trivial.
Next, we note that since the group is Abelian all of its non-projective irreducible representations are 1-dimensional,
and therefore the representation g → u∗β (g) ⊗ eis(g) uγ (g) can be decomposed to 1-dimensional irreducible represen-
tations of G. Let the normalized vector |Θ(κ)i be a 1-dimensional subspace on which g → u∗β (g) ⊗ eis(g) uγ (g) acts
irreducibly, i.e.
h i
∀g ∈ G : u∗β (g) ⊗ eis(g) uγ (g) |Θ(κ)i = eiκ(g) |Θ(κ)i , (A4)

for some 1-dimensional representation eiκ(g) . Then, using the fact that the representations β and γ are irreducible,
it follows from the Schur’s lemma that |Θ(κ)i should be a maximally entangled state, and its corresponding reduced
state on both subsystems should be proportional to the identity operator (To see this consider the reduced density
operator corresponding to state |Θ(κ)i on one subsystem, and then use Eq.(A4) to show that this reduced state should
commute with an irreducible representation. Then, using Schur’s lemma we find that this density operator should be
proportional to the identity operator, which means state |Θ(κ)i should be maximally entangled.).
This implies that the dimension of the irreducible representations β and γ are the same. We denote this dimension
by d. Furthermore, since |Θ(κ)i is a maximally entangled normalized vector, there exists a unitary V such that
d
1 X
|Θ(κ)i = (V ⊗ I)|ψ (+) i = (V ⊗ I) √ |iii , (A5)
d i=1
Pd
where |ψ (+) i ≡ √1 |iii, and I is the identity operator acting on the d-dimensional space. Then Eq.(A4) implies
d i=1

 
(V † u∗β (g)V ) ⊗ uγ (g) |ψ (+) i = ei(κ(g)−s(g)) |ψ (+) i . (A6)

Next, we use the fact that for any operator B we have (I ⊗ B)|ψ (+) i = (B T ⊗ I)|ψ (+) i, where B T is the transpose of
B. It follows that
 † ∗

V uβ (g)V uTγ (g) ⊗ I |ψ (+) i = ei(κ(g)−s(g)) |ψ (+) i . (A7)

This implies that V † u∗β (g)V uTγ (g) = ei(κ(g)−s(g)) I, and therefore

∀g ∈ G : uγ (g) = ei(κ(g)−s(g)) V T uβ (g)V ∗ . (A8)

Choosing W = V T , and eir(g) = ei(κ(g)−s(g)) we obtain Eq.(A1).