Entanglement detection length of multipartite quantum states

Fei Shi,1 Lin Chen,2, 3, ∗ Giulio Chiribella,1, 4, 5, † and Qi Zhao1, ‡

1
QICI Quantum Information and Computation Initiative, Department of Computer Science,
The University of Hong Kong, Pokfulam Road, Hong Kong
2
LMIB(Beihang University), Ministry of Education,
and School of Mathematical Sciences, Beihang University, Beijing 100191, China
3
International Research Institute for Multidisciplinary Science, Beihang University, Beijing 100191, China
4
Department of Computer Science, Parks Road, Oxford, OX1 3QD, United Kingdom
5
Perimeter Institute for Theoretical Physics, Waterloo, Ontario N2L 2Y5, Canada
Multipartite entanglement is a crucial resource for quantum computing, communication, and
metrology. However, detecting this resource can be challenging: for genuine multipartite entan-
glement it may require global measurements that are hard to implement experimentally. In this
study, we introduce the concept of entanglement detection length, defined as the minimum length
arXiv:2401.03367v1 [quant-ph] 7 Jan 2024

of observables required to detect genuine multipartite entanglement. We characterize the entangle-

ment detection length for various types of genuinely entangled states, including GHZ-like states,
Dicke states, and graph states. We also present a classification of genuinely entangled states based
on entanglement detection length. Furthermore, we demonstrate that the entanglement detection
length differs from the minimum length of observables needed to uniquely determine a genuinely
entangled state. Our findings are valuable for minimizing the number of observables that must be
measured in entanglement detection.

Multipartite entanglement plays an important role in
quantum information processing tasks. In addition, it
has a fundamental interest in the study of quantum non-
locality. Genuine multipartite entanglement (GME) em-
bodies global entanglement and is acknowledged as the
most potent form of entanglement. The applications of
GME span a diverse range of areas, including quantum
communication [1], quantum computing [2] and quantum
metrology [3]. Genuinely entangled states of up to 51 su-
perconducting qubits have been generated [4], and even
larger sizes are likely to become available as experimental
platforms advance.
A popular classification of genuinely entangled pure
states is based on stochastic local operations and classi-
cal communication (SLOCC) [5, 6]. For three-qubit gen-
uinely entangled pure states, there are only two SLOCC
classes, represented by the GHZ state and the W state.
However, for n-qubit genuinely entangled pure states
with n ≥ 4, infinitely many of SLOCC classes arise [6–8],
which makes this classification too fine-grinding. Mean-
while, several GME measures have been proposed, such
as genuine multipartite concurrence [9], concurrence fill
[10], and the geometric mean of bipartite concurrences
[11]. These GME measures can be employed to quan-
tify GME; however, their behaviors vary for distinct gen-
uinely entangled pure states, resulting in different opti-
mal entanglement states. Notably, some of these mea-
sures can increase under SLOCC or LOCC [11, 12].
Besides the classification and quantification of GME in
theory, an important problem is the detection of GME
in the laboratory. The straightforward approach is to
use global observables, which act non-trivially on a large
number of subsystems. However, the expectation values
of such observables are often difficult to access experi-
mentally [13, 14]. Hence, it is important to devise detec-
tion strategies that use local observables acting on a small
number of neighboring systems. Examples of such strate-
gies were provided in Refs. [13–15], which showed exam-
ples of genuinely entangled states that can be detected by
two-body observables. Despite these advances, the prob-
lem of detecting multipartite entanglement through few-
body observables still lacks a systematic treatment, and
several basic questions remain unanswered. For exam-
ple, given an entangled state of a one-dimensional chain
of quantum systems, one can ask what is the minimum
length of the observables needed to detect the GME (if
any) of that state. Determining the minimum, hereafter
called entanglement detection length, is useful for the de-
sign of entanglement detection protocols, and can poten-
tially reveal an aspect of multipartite entanglement that
was not captured by previous quantities.
An analogue of the entanglement detection length
emerges in the study of quantum state tomography via
reduced density matrices [16–22]. In this context, an
important question is what is the minimum observable
length k needed to uniquely determine a global state
from its k-body marginals. We call such a minimum the
state determination length. In an entanglement detection
task, we only need to detect a genuinely entangled state
instead of uniquely identifying a state. Thus, state deter-
mination length serves as an upper bound for the entan-
glement detection length. An interesting open question
is whether the entanglement detection length is the same
as the state determination length for the same genuinely
entangled state.
This paper embarks on an investigation into these
pressing issues by characterizing the entanglement de-
tection length for various types of genuinely entangled
states. For example, we show that an n-qubit GHZ
state has an entanglement detection length of n, while an

n-qubit Dicke state exhibits an entanglement detection
length of 2. We also explore the entanglement detection
length for other widely used types of genuinely entangled
states, such as graph states. We show the state deter-
mination length for several types of genuinely entangled
states and find that the entanglement detection length
is different from the state determination length by giv-
ing counter-examples with the various gaps. Moreover,
we classify the genuinely entangled states according to
entanglement detection length, with total n − 1 classes,
which is different from the classification of genuinely en-
tangled states based on SLOCC. Our entanglement de-
tection length is inconsistent with some existing GME
measures [9–11]. Our work gives the fundamental limits
for the length of observables in entanglement detection,
which can enhance the understanding and fundamental
exploration of multipartite entanglement.
Entanglement detection length.—Let D be the set of
all density matrices on the tensor product Hilbert space
H := ⊗ni=1 Hi . A state ρ is genuinely entangled if it is
not aP biseparable state, that
P is if it cannot be written as
ρ = i pi |ψ i ⟩⟨ψ i |, with i pi = 1, pi ≥ 0 and |ψi ⟩ =
|αi ⟩Sj1 ⊗ |βi ⟩Sj2 where Sj1 , Sj2 ⊆ [n] := {1, 2, . . . , n},
are non-empty N disjoint subsets forming
Na partition of [n],
and |αi ⟩Sj1 ∈ i∈Sj1 Hi , |β i ⟩Sj ∈ i∈Sj2 Hi are two
2
unit vectors.
Given a quantum state ρ ∈ D and a collection of sub-
sets S = {S1 , . . . , Sk }, with Sj ⊆ [n], ∀j ∈ {1, . . . , k}, we
define the compatibility set C(ρ, S) as the set of all den-
sity matrices σ ∈ D that have the same marginals as ρ
on all the subsystems associated to sets in S: in formula,
 The above lemma implies a bound on entanglement
C(ρ, S) := σ ∈ D σS = ρS , ∀S ∈ S . (1)
Note that the compatibility set is a convex set [20].
If the compatibility set C(ρ, S) contains only genuinely
entangled states, then GME can be guaranteed by the
specification of the reduced density matrices (ρS )S∈S . In
turn, the reduced density matrices can be in principle de-
termined by performing measurements only on the sub-
systems labeled by S. This observation motivates the
following definition:
Definition 1: We say that a state ρ ∈ D
has S-detectable genuine multipartite entanglement (S-
detectable GME, for short) if the compatibility set C(ρ, S)
contains only genuinely entangled states. When this is
the case, we say that S detects ρ’s GME.
In the following, we will denote l(S) := maxS∈S |S| as
the maximum size of a subset in S. The entanglement
detection length of the state ρ is then defined as
 we investigate the entanglement detection length for sev-
l(ρ) := min l(S) S detects ρ’s GME . (2)
S
If a genuinely entangled state has entanglement detection
length l, then its GME cannot be detected by using only
b-body measurements with b < l: every experimental
2
scheme that certifies the presence of GME must include
at least one l-body observable.
Conversely, the set of all l-body marginals of the state
is sufficient to certify the presence of GME: every state
with the same l-body marginals must necessarily be gen-
uinely entangled. As a consequence, GME can be guar-
n
anteed by doing a tomography of at most reduced
l
density matrices. It is interesting to consider the mini-
mum number of l-body marginals needed to detect GME.
We denote the minimum as

m(ρ) := min |S| S detects ρ’s GME . (3)
S: |S|=l(ρ) ,∀S∈S
We now provide a few graph-theoretic observations.
Given a collection of subsets S = {S1 , . . . , Sk } , Sj ⊆
[n] , ∀j ∈ {1, . . . , k}, we consider the hypergraph G =
([n], S) with vertices in [n] and hyperedges in S. A first
elementary observation is that a set S can detect GME
only if the corresponding hypergraph G = ([n], S) is con-
nected:
Lemma 1: If the hypergraph G = ([n], S) is not con-
nected, then, for every state ρ ∈ D, the compatibility set
C(ρ, S) contains a product state.
Proof.—– If the hypergraph G = ([n], S) is not con-
nected, then its vertex set [n] can be partitioned into two
non-empty disjoint sets X and Y such that no hyper-
edge connects a vertex in X to a vertex in Y . Then, for
every state ρ, the compatibility set C(ρ, S) contains the
product state ρX ⊗ ρY . ■
detection length of a genuinely entangled state,
2 ≤ l(ρ) ≤ n. (4)
The above lemma also implies a bound on the quantity
m(ρ) defined in Eq. (3). This quantity is defined in
terms of a minimization over all collections of subsets
S = {S1 , . . . , Sk } with the property |Sj | = l(ρ), ∀j ∈
{1, . . . , k}. This property amounts to the condition that
the hypergraph G = ([n], S) is l(ρ)-uniform, meaning
that all its edges have cardinality l(ρ). Since the mini-
mum number of edges for la connectedm l-uniform hyper-
graph with n vertices is n−1 l−1 (see Appendix A), we
obtain the bound
   
n−1 n
≤ m(ρ) ≤ . (5)
l(ρ) − 1 l(ρ)
Evaluation of the entanglement detection length. Here
eral paradigmatic families of genuinely entangled states,
namely GHZ-like states, Dicke states, and graph states.
An n-qubit GHZ-like state | GHZ∗n ⟩ can be written as
|GHZ∗n ⟩ = a|α1 α2 · · · αn ⟩ + b|β1 β2 · · · βn ⟩, (6)

Ring states
𝟐 𝟑 𝟒 … 𝒏− 𝟏
Dicke states
Cluster states
𝑘-Uniform graph states with 𝑘 ≥ 3
GHZ-like states
FIG. 1: Entanglement detection lengths of different
genuinely entangled states. The entanglement detection
length of a genuinely entangled state is the minimum
observable length needed to detect the GME of this
state. Genuinely entangled states can be classified into
n − 1 classes based on entanglement detection length.
where |a|2 + |b|2 = 1, ab ̸= 0, ⟨αi |αi ⟩ = ⟨βi |βi ⟩ = 1,
and ⟨αi |βi ⟩ = 0 for 1 ≤ i ≤ n. Let S be the set
of all (n − 1)-subsets of [n]. Since the compatibility
set C(| GHZ∗n ⟩, S) contains a fully separable state |a|2 ·
|α1 α2 · · · αn ⟩⟨α1 α2 · · · αn | + |b|2 · |β1 β2 · · · βn ⟩⟨β1 β2 · · · βn |,
we have l(| GHZ∗n ⟩) = n, and m(| GHZ∗n ⟩) = 1. If an
n-qubit genuinely entangled pure state |ψ⟩ is not a GHZ-
like state, then the compatibility set C(| GHZ∗n ⟩, S) con-
tains only |ψ⟩ [23], which means that S detects |ψ⟩’s
GME. So we can give a necessary and sufficient condi-
tion for l(|ψ⟩) = n.
Proposition 1: For an n-qubit genuinely entangled
pure state |ψ⟩, l(|ψ⟩) = n if and only if |ψ⟩ is a GHZ-like
state.
The n-qubit Dicke state can be written as
1 X
|Dnk ⟩ = q
|i1 , i2 , · · · , in ⟩,
n that its entanglement detection length is 4.
k (i1 ,i2 ,··· ,in )∈Zn
2,
wt(i1 ,i2 ,··· ,in )=k
where 1 ≤ k ≤ n − 1, n ≥ 3, and wt(i1 , i2 , · · · , in ) is
the number of nonzero ij for 1 ≤ j ≤ n. Specially, when
k = 1, the Dicke state is the W state |Wn ⟩.
Theorem 1: For all 1 ≤ k ≤ n−1 and n ≥ 3, S detects
the Dicke state |Dnk ⟩’s GME if and only if the hypergraph
G = ([n], S) is connected. Moreover, l(|Dnk ⟩) = 2, and
m(|Dnk ⟩) = n − 1.
The proof of Theorem 1 is given in Appendix C. Con-
sider an ordinary graph G = (V, E) with V = [n]. The
3
neighborhood NG (i) of a vertex i is the set of all the
vertices adjacent to i. The degree degG (i) of a ver-
tex i is the number of vertices in NG (i). We denote
∆(G) as the maximum degree of a vertex in the graph
G. For every N vertex i, we can define a Pauli string
gi = Xi ⊗ j∈NG (i) Zj . The graph state |Gn ⟩ is the
unique simultaneous +1-eigenvalue eigenstate of the n
operators {gi }ni=1 , i.e. gi |Gn ⟩ = |Gn ⟩ for 1 ≤ i ≤ n. The
graph state is genuinely entangled [24] when the graph
𝒏
is connected. Graph states can be also used to construct
k-uniform states [25, 26]. An n-qubit pure state is a k-
uniform state if each k-body marginal  n  is proportional to
the identity, and it must have k ≤ 2 [27, 28]. Specially,
a n2 -uniform state is also called an absolutely maxi-

mally entangled (AME) state, and n-qubit AME states
exist only for n = 2, 3, 5, 6 [29].
Proposition 2: Let G be a connected graph of n-
vertices (n ≥ 3), then the entanglement detection length
of the graph state |Gn ⟩ satisfy
3 ≤ l(|Gn ⟩) ≤ ∆(G) + 1. (8)
(k)
Specially, if the graph state is a k-uniform state |Gn ⟩,
then
k + 1 ≤ l(|G(k)
n ⟩) ≤ ∆(G) + 1. (9)
Proof.— When G is a connected graph, the compat-
ibility set C(|Gn ⟩, S) contains a fully separable state,
where S consists of all the 2-subsets of [n] [30], which
implies l(|Gn ⟩) ≥ 3. The compatibility set C(|Gn ⟩, {{i} ∪
NG (i)}ni=1 ) contains only the graph state [31], which
means that l(|Gn ⟩) ≤ ∆(G) + 1. Since compatibility set
C(|Gn ⟩, S ′ ) contains a fully separable state, i.e. 21n IH ,
where S ′ consists of all the k-subsets of [n], we have
(k)
l(|Gn ⟩ ≥ k + 1. ■
If G is a line with n ≥ 3 vertices, then the graph state
is the cluster state with l(Clustern ) = 3; if G is a circle
with n ≥ 3 vertices, then the graph state is the ring state
with l(Ringn ) = 3. From Eq. (5) and the results in [31],
 n−1
we know that 2 ≤ m(Clustern ) ≤ n − 2 for n ≥ 3,
and n−1
 
2 ≤ m(Ring n ) ≤ n for n ≥ 3. The 6-qubit
AME state can be obtained from a 3-regular graph [25],
where each vertex has degree 3, then we can determine
(7)
See Fig. 1 for a summary of the entanglement detec-
tion lengths of different genuinely entangled pure states.
From Fig. 1, we can classify genuinely entangled states
based on entanglement detection length. There are only
n − 1 classes of n-qubit genuinely entangled states in this
classification. Note that this classification is different
from the classification of genuinely entangled states based
on SLOCC. For example, there always exists a pure state
|ψ⟩ which is SLOCC equivalent to a GHZ-like state, and
it is not a GHZ-like state. By Proposition 1, l(|ψ⟩) < n,
and l(| GHZ∗n ⟩) = n, which means that the entanglement
 4

TABLE I: Comparison between entanglement detection length l(|ψ⟩) and state determination length L(|ψ⟩) of
diffierent genuinely entangled states. m(|ψ⟩) (resp. M (|ψ⟩)) is the minimum number of the l(|ψ⟩)-body (resp.
L(|ψ⟩)-body) marginals that can be used to detect |ψ⟩’s GME (uniquely determine |ψ⟩). The
gap(|ψ⟩) := L(|ψ⟩) − l(|ψ⟩), and [a, b] means the lower bound of m(|ψ⟩) (resp. M (|ψ⟩)) is a and the upper bound of
m(|ψ⟩) (resp. M (|ψ⟩)) is b.

genuinely entangled state |ψ⟩ l(|ψ⟩) m(|ψ⟩) L(|ψ⟩) M (|ψ⟩) gap(|ψ⟩)

GHZ-like state | GHZ∗n ⟩, n ≥ 3 n 1 n 1 0

Dicke state |Dnk ⟩, 1 ≤ k ≤ n − 1, n ≥ 3 2 n−1 2 n−1 0
Cluster state |Clustern ⟩, n ≥ 3 3 [ n−1
2
,n − 2] 3 [ n−1
2
,n − 2] 0
Ring state |Ringn ⟩, n ≥ 3 3 [ n−1
2
, n] 3 [ n−1
2
, n] 0
|ψ⟩ = a|Wn ⟩ + b|11 · · · 1⟩, ab ̸= 0, n ≥ 4, |ψ⟩⟨ψ|{1,2} is NPT 2 n−1 n−1 2 n−3
 n−1 
|ψ⟩ = a|Wn ⟩ + b|11 · · · 1⟩, ab ̸= 0, n ≥ 4, |ψ⟩⟨ψ|{1,2} is PPT 3 2
n−1 2 n−4

detection length of a genuinely entangled state could be
changed under SLOCC. Let S be the set of all 2-subsets
of [n], and |Wn ⟩ be a state that is SLOCC equivalent to
the W state | Wn ⟩, then the compatibility set C(|Wn ⟩, S)
contains only |Wn ⟩ [32]. Therefore, the entanglement de-
tection length of the W state is invariant under SLOCC.
Differences to state determination length.—
Reconstruction of a quantum state via its reduced
density matrices is an efficient tool for quantum state
tomography, which attracted a lot of attention [16–22].
Given an n-qubit state ρ, and a collection of subsets S, if
the compatibility set C(ρ, S) contains only ρ, then we say
that S uniquely determines ρ. In this situation, one only
needs to measure the marginals on all the subsystems
associated to sets in S to reconstruct ρ, which could
reduce an exponential number of measurements [33].
In comparison to entanglement detection length l(ρ),
and minimum number of l(ρ)-body marginals needed to
detect ρ’s GME m(ρ), we can also define L(ρ) as the state
determination length, and M (ρ) as the minimum number
of L(ρ)-body marginals needed to uniquely determine ρ.
See Appendix D for the detailed definitions. From the
definitions of entanglement detection length and state
determination length, it is clear that
2 ≤ l(ρ) ≤ L(ρ) ≤ n,
for an n-qubit genuinely entangled state ρ. Then we can
define the gap between these two lengths
gap(ρ) = L(ρ) − l(ρ).
From the process of calculating the entanglement de-
tection lengths for GHZ-like states, Dicke states, cluster
states and ring states, there is no gap between entangle-
ment detection length and state determination length for

S uniquely
these states. In [32], the authors showed that
determines |Dnk ⟩, where S consists of n−1k numbers of
(k + 1)-subsets of [n], for all 1 ≤ k ≤ n − 1, and n ≥ 3.
From the proof of Theorem 1, we show that S ′ uniquely
determines |Dnk ⟩, where S ′ consists of n − 1 numbers of
2-subsets of [n] for all 1 ≤ k ≤ n − 1, and n ≥ 3, which
greatly improve the result in [32].
For the same discussion as Proposition 1, if |ψ⟩ is a
genuinely entangled pure state, then L(|ψ⟩) = n if and
only if |ψ⟩ is a GHZ-like state. This implies that
0 ≤ gap(|ψ⟩) ≤ n − 3, (12)
for an n-qubit genuinely entangled state |ψ⟩. Specially,
for each three-qubit genuinely entangled pure state |ψ⟩,
gap(|ψ⟩) = 0. More specifically, l(|ψ⟩) = L(|ψ⟩) = 3 if
|ψ⟩ = | GHZ∗3 ⟩. Otherwise, l(|ψ⟩) = L(|ψ⟩) = 2.
Is there a genuinely entangled pure state with a distinct
gap between its entanglement detection length and state
determination length? Next, we answer this question by
showing that there is not only a gap between these two
lengths, but a maximum gap can be reached. If ρ is
positive (non-positive) under partial transpose, then it
is said that ρ is positive partial transpose (PPT) (non-
positive partial transpose (NPT)). We denote |ψ⟩⟨ψ|S =
Tr[n]\S |ψ⟩⟨ψ|.
Theorem 2: Let
|ψ⟩ = a|Wn ⟩ + b|11 · · · 1⟩, (13)
(10)
where ab ̸= 0, |a|2 + |b|2 = 1, and n ≥ 4. Then L(|ψ⟩) =
n − 1, M (|ψ⟩) = 2. Moreover,
(i) If |ψ⟩⟨ψ|{1,2} is NPT, then
(11)
l(|ψ⟩) = 2, m(|ψ⟩) = n − 1, gap(|ψ⟩) = n − 3.
(ii) If |ψ⟩⟨ψ|{1,2} is PPT, then

n−1

l(|ψ⟩) = 3, m(|ψ⟩) = , gap(|ψ⟩) = n − 4.
2
The proof of Theorem
√ 2 is given
√ in Appendix E. For
example, when |ψ⟩ = 0.9|W4 ⟩+ 0.1|1111⟩, gap(|ψ⟩) =

√ √
1; when |ψ⟩ = √ 0.94|W5 ⟩ + √ 0.06|11111⟩, gap(|ψ⟩) = 2;
and when |ψ⟩ = 0.97|W6 ⟩ + 0.03|111111⟩, gap(|ψ⟩) =
3. Note that we can also classify genuinely entangled
states based on state determination length. However,
Theorem 2 shows that this classification is different from
the classification of genuinely entangled states based on
entanglement detection length.
Our entanglement detection length can be also used to
quantify GME to some extent. For example, the GHZ-
like states are “more entangled” than the Dicke states
in terms of entanglement detection. Although there are
several GME measures that have been proposed, such as
genuine multipartite concurrence [9], concurrence fill [10],
and the geometric mean of bipartite concurrences [11],
our entanglement detection length is inconsistent with
these GME measures. For instance, the geometric mean
of bipartite concurrences reaches its maximum value for
AME states [10], while our entanglement detection length
achieves its maximum value for GHZ-like states.
Conclusions and discussion.—We have introduced the
entanglement detection length, which is the minimum ob-
servable length for detecting genuinely entangled states,
and see Table I for a summary. Our entanglement de-
tection length can be used to classify and quantify gen-
uinely entangled states. Moreover, our entanglement de-
tection length is also different from the state determina-
tion length with various gaps. Based on entanglement de-
tection length, we can detect genuinely entangled states
by tomography. We believe our method is useful for de-
tecting GME experimentally. Our approaches can be ex-
panded to encompass qudit systems as well. In addition,
while our definition is applicable to mixed states, exam-
ining the entanglement detection length of genuinely en-
tangled mixed states, such as the Werner state, would be
intriguing. Although we provide the entanglement de-
tection length for various types of genuinely entangled
states, constructing an entanglement witness that can
saturate the entanglement detection length remains an
unknown process.
We thank Xiao Yuan, Xiongfeng Ma, You Zhou, and
Xiande Zhang for their helpful discussion and sugges-
tions. F. S. and G. C. acknowledges funding from the
Hong Kong Research Grant Council through Grants No.
17300918 and No. 17307520, and through the Senior
Research Fellowship Scheme SRFS2021-7S02. This pub-
lication was made possible through the support of the
ID# 62312 grant from the John Templeton Foundation,
as part of the ‘The Quantum Information Structure of
Spacetime’ Project (QISS). The opinions expressed in
this project are those of the authors and do not necessar-
ily reflect the views of the John Templeton Foundation.
Research at the Perimeter Institute is supported by the
Government of Canada through the Department of In-
novation, Science and Economic Development Canada
and by the Province of Ontario through the Ministry
of Research, Innovation and Science. L. C. was sup-
5
ported by the NNSF of China (Grant No. 11871089),
and the Fundamental Research Funds for the Central
Universities (Grant Nos. KG12040501, ZG216S1810 and
ZG226S18C1). Q.Z. acknowledges funding from HKU
Seed Fund for Basic Research for New Staff via Project
2201100596, Guangdong Natural Science Fund—General
Programme via Project 2023A1515012185, National Nat-
ural Science Foundation of China (NSFC) Young Scien-
tists Fund via Project 12305030, 27300823, Hong Kong
Research Grant Council (RGC) via No. 27300823,
and NSFC/RGC Joint Research Scheme via Project
N HKU718/23.
∗
linchen@buaa.edu.cn
†
giulio@cs.hku.hk
‡
zhaoqi@cs.hku.hk
[1] A. Karlsson and M. Bourennane, “Quantum teleporta-
tion using three-particle entanglement,” Physical Review
A, vol. 58, no. 6, p. 4394, 1998.
[2] A. Steane, “Quantum computing,” Reports on Progress
in Physics, vol. 61, no. 2, p. 117, 1998.
[3] A. W. Chin, S. F. Huelga, and M. B. Plenio, “Quantum
metrology in non-markovian environments,” Physical re-
view letters, vol. 109, no. 23, p. 233601, 2012.
[4] S. Cao, B. Wu, F. Chen, M. Gong, Y. Wu, Y. Ye, C. Zha,
H. Qian, C. Ying, S. Guo et al., “Generation of genuine
entanglement up to 51 superconducting qubits,” Nature,
vol. 619, no. 7971, pp. 738–742, 2023.
[5] J. Eisert and H. J. Briegel, “Schmidt measure as a tool
for quantifying multiparticle entanglement,” Physical Re-
view A, vol. 64, no. 2, p. 022306, 2001.
[6] F. Verstraete, J. Dehaene, B. De Moor, and H. Ver-
schelde, “Four qubits can be entangled in nine different
ways,” Physical Review A, vol. 65, no. 5, p. 052112, 2002.
[7] W. Dür, G. Vidal, and J. I. Cirac, “Three qubits can be
entangled in two inequivalent ways,” Physical Review A,
vol. 62, no. 6, p. 062314, 2000.
[8] H. Dietrich, W. A. De Graaf, A. Marrani, and M. Origlia,
“Classification of four qubit states and their stabilisers
under slocc operations,” Journal of Physics A: Mathe-
matical and Theoretical, vol. 55, no. 9, p. 095302, 2022.
[9] Z.-H. Ma, Z.-H. Chen, J.-L. Chen, C. Spengler,
A. Gabriel, and M. Huber, “Measure of genuine mul-
tipartite entanglement with computable lower bounds,”
Physical Review A, vol. 83, no. 6, p. 062325, 2011.
[10] Y. Li and J. Shang, “Geometric mean of bipartite concur-
rences as a genuine multipartite entanglement measure,”
Physical Review Research, vol. 4, no. 2, p. 023059, 2022.
[11] S. Xie and J. H. Eberly, “Triangle measure of tripartite
entanglement,” Physical Review Letters, vol. 127, no. 4,
p. 040403, 2021.
[12] X. Ge, L. Liu, and S. Cheng, “Tripartite entanglement
measure under local operations and classical communi-
cation,” Physical Review A, vol. 107, no. 3, p. 032405,
2023.
[13] J. Tura, R. Augusiak, A. B. Sainz, T. Vértesi, M. Lewen-
stein, and A. Acı́n, “Detecting nonlocality in many-body
quantum states,” Science, vol. 344, no. 6189, pp. 1256–
1258, 2014.
 6

[14] F. Baccari, D. Cavalcanti, P. Wittek, and A. Acı́n, “Effi- classical communication,” Physical Review A, vol. 84,
cient device-independent entanglement detection for mul- no. 5, p. 052331, 2011.
tipartite systems,” Physical Review X, vol. 7, no. 2, p. [33] M. Cramer, M. B. Plenio, S. T. Flammia, R. Somma,
021042, 2017. D. Gross, S. D. Bartlett, O. Landon-Cardinal, D. Poulin,
[15] M. Markiewicz, W. Laskowski, T. Paterek, and and Y.-K. Liu, “Efficient quantum state tomography,”
M. Żukowski, “Detecting genuine multipartite entangle- Nature communications, vol. 1, no. 1, p. 149, 2010.
ment of pure states with bipartite correlations,” Physical [34] T. Ichikawa, T. Sasaki, I. Tsutsui, and N. Yonezawa, “Ex-
Review A, vol. 87, no. 3, p. 034301, 2013. change symmetry and multipartite entanglement,” Phys-
[16] N. Linden, S. Popescu, and W. Wootters, “Almost every ical Review A, vol. 78, no. 5, p. 052105, 2008.
pure state of three qubits is completely determined by its
two-particle reduced density matrices,” Physical review
letters, vol. 89, no. 20, p. 207901, 2002.
[17] N. Linden and W. Wootters, “The parts determine the
whole in a generic pure quantum state,” Physical review
letters, vol. 89, no. 27, p. 277906, 2002.
[18] L. Diósi, “Three-party pure quantum states are deter-
mined by two two-party reduced states,” Physical Review
A, vol. 70, no. 1, p. 010302, 2004.
[19] J. Chen, Z. Ji, M. B. Ruskai, B. Zeng, and D.-L. Zhou,
“Comment on some results of erdahl and the convex
structure of reduced density matrices,” Journal of Math-
ematical Physics, vol. 53, no. 7, 2012.
[20] J. Chen, Z. Ji, D. Kribs, Z. Wei, and B. Zeng, “Ground-
state spaces of frustration-free hamiltonians,” Journal of
mathematical physics, vol. 53, no. 10, p. 102201, 2012.
[21] J. Chen, H. Dawkins, Z. Ji, N. Johnston, D. Kribs,
F. Shultz, and B. Zeng, “Uniqueness of quantum states
compatible with given measurement results,” Physical
Review A, vol. 88, no. 1, p. 012109, 2013.
[22] T. Xin, D. Lu, J. Klassen, N. Yu, Z. Ji, J. Chen, X. Ma,
G. Long, B. Zeng, and R. Laflamme, “Quantum state to-
mography via reduced density matrices,” Physical review
letters, vol. 118, no. 2, p. 020401, 2017.
[23] S. N. Walck and D. W. Lyons, “Only n-qubit greenberger-
horne-zeilinger states contain n-partite information,”
Physical Review A, vol. 79, no. 3, p. 032326, 2009.
[24] M. Hein, J. Eisert, and H. J. Briegel, “Multiparty en-
tanglement in graph states,” Physical Review A, vol. 69,
no. 6, p. 062311, 2004.
[25] W. Helwig, “Absolutely maximally entangled qudit
graph states,” arXiv preprint arXiv:1306.2879, 2013.
[26] S. Sudevan and S. Das, “n-qubit states with maximum
entanglement across all bipartitions: A graph state ap-
proach,” arXiv preprint arXiv:2201.05622, 2022.
[27] A. J. Scott, “Multipartite entanglement, quantum-error-
correcting codes, and entangling power of quantum evo-
lutions,” Physical Review A, vol. 69, no. 5, p. 052330,
2004.
[28] D. Goyeneche and K. Życzkowski, “Genuinely multipar-
tite entangled states and orthogonal arrays,” Physical re-
view A, vol. 90, no. 2, p. 022316, 2014.
[29] F. Huber, O. Gühne, and J. Siewert, “Absolutely max-
imally entangled states of seven qubits do not exist,”
Physical review letters, vol. 118, no. 20, p. 200502, 2017.
[30] O. Gittsovich, P. Hyllus, and O. Gühne, “Multiparti-
cle covariance matrices and the impossibility of detect-
ing graph-state entanglement with two-particle correla-
tions,” Physical Review A, vol. 82, no. 3, p. 032306, 2010.
[31] X. Wu, Y.-h. Yang, Y.-k. Wang, Q.-y. Wen, S.-j. Qin,
and F. Gao, “Determination of stabilizer states,” Physi-
cal Review A, vol. 92, no. 1, p. 012305, 2015.
[32] S. Rana and P. Parashar, “Optimal reducibility of all w
states equivalent under stochastic local operations and
 7

Appendix A: The minimum number of hyperedges for a connected l-uniform hypergraph

FIG. 2: Three 3-uniform connected hypergraphs. The left hypergraph has one hyperedge; the center hypergraph has
two hyperedges; the right hypergraph has three hyperedges.

For a l-uniform connected hypergraph, 1 hyperedge covers l vertices; 2 hyperedges cover at most 2l − 1 vertices; 3
hyperedges cover at most 3l − 2 vertices (See Fig. 2). Then k hyperedges cover at most kl − k + 1 vertices.
l m For a
n−1
l-uniform connected hypergraph with n vertices and k hyperedges, it must have kl − k + 1 ≥ n, i.e. k ≥ l−1 . Thus,
l m
the minimum number of hyperedges for a l-uniform connected hypergraph with n vertices is n−1 l−1 .

Appendix B: Some properties for symmetric states

A symmetric (bosonic) space HSym is a subspace of H = ⊗ni=1 Hi , where each pure state is invariant under the
swap of any two subsystems, i.e.

HSym := {|ψ⟩ ∈ H : Swap(i,j) |ψ⟩ = |ψ⟩, ∀1 ≤ i, j ≤ n}, (14)

where Swap(i,j) is a swap operator that swaps i-th and j-th subsystems. Each pure state of HSym is called a symmetric
pure state. Note that Dim HSym = n + 1, and HSym has an orthogonal basis {|Dnk ⟩}nk=0 consisting of Dicke states,
where
1 X
|Dnk ⟩ = q
|(i1 , i2 , . . . , in )⟩, (15)
n
k (i1 ,i2 ,...,in )∈ZN
2 ,wt(i1 ,i2 ,...,in )=k

and wt(i1 , i2 , · · · , in ) is the number of nonzero ij for 1 ≤ j ≤ n. A symmetric pure state |ψ⟩ can be written as
n
X
|ψ⟩ = ak |Dnk ⟩, (16)
k=0
Pn
where k=0 |ak |2 = 1.
For a state ρ, we denote R(ρ) as the range of ρ. Next, we introduce the concept of symmetric mixed states.

Definition 1 An n-qubit state ρ is a symmetric state if R(ρ) ⊆ HSym . In other words, ρ = Swap(i,j) ρ for all
1 ≤ i, j ≤ n.
P
A symmetric state can be written as ρ = i pi |ψi ⟩⟨ψi |, where |ψi ⟩ ∈ HSym . Furthermore, a symmetric state ρ can
also be written as
n
X
ρ= ai,j |Dni ⟩⟨Dnj |, (17)
i,j=0

where A = (ai,j )0≤i,j≤n is positive and Tr(A) = 1. A symmetric state ρ is either a fully separable
P state or a genuinely
entangled state [34]. When ρ is a fully separable symmetric state, it can be written as ρ = i pi |αi αi · · · αi ⟩⟨αi αi · · · αi |
[34]. Next, we give some properties for symmetric states.

Lemma 1 Assume ρ is an n-qubit symmetric state, then the reduced density matrix ρS is also a symmetric state for
each nonempty subset S ⊆ [n]. Furthermore, ρ{1,2,...,k} = ρ{i1 ,i2 ,··· ,ik } for all 1 ≤ k ≤ n and {i1 , i2 , · · · , ik } ⊆ [n].
 8

Proof. Let S = {i1 , i2 , · · · , ik } ⊆ [n], then

Swap(is ,it ) ρS = Swap(is ,it ) TrS (ρ) = TrS (Swap(is ,it ) ρ) = TrS ρ = ρS , ∀1 ≤ s, t ≤ k. (18)
Thus ρS must be a symmetric state.
If |ψ⟩ is a symmetric pure state, then |ψ⟩⟨ψ|{1,2,...,k} P = |ψ⟩⟨ψ|{i1 ,i2 ,··· ,ik } for all 1 ≤ k ≤ n and {i1 , i2 , · · · , ik } ⊆ [n].
Since a symmetric mixed state can be written as ρ = i pi |ψi ⟩⟨ψi |, where |ψi ⟩ ∈ HSym , we obtain that ρ{1,2,...,k} =
ρ{i1 ,i2 ,··· ,ik } for all 1 ≤ k ≤ n and {i1 , i2 , · · · , ik } ⊆ [n].
Lemma 2 Assume ρ is an n-qubit state and S is a collection of subsets, with S ⊆ [n], ∀S ∈ S. If ρS is a symmetric
state for all S ∈ S, and the hypergraph G = ([n], S) is connected, then ρ is also a symmetric state.
Proof. Firstly, we need to show that if ρ{i,j} is a symmetric state, then Swap(i,j) ρ = ρ.
We consider the n-qubit pure state |ψ⟩. By Schmidt decomposition, |ψ⟩ can be written as
X
|ψ⟩ = λk |αk ⟩{i,j} |βk ⟩{i,j} ,
k

λ2k = 1, and the states {|αk ⟩{i,j} } ({|βk ⟩{i,j} }) are mutually orthogonal. If
P
where λk ≥ 0, k
X
|ψ⟩⟨ψ|{i,j} = λ2k |αk ⟩{i,j} ⟨αk | (19)
k

is a symmetric state, then |αk ⟩{i,j} is a symmetric pure state for each k, i.e. Swap(i,j) |αk ⟩{i,j} = |αk ⟩{i,j} for each k.
It implies that Swap(i,j) |ψ⟩ = |ψ⟩. P P
Next, we consider the n-qubit mixed state ρ = k pk |ψk ⟩⟨ψk |. If ρ{i,j} = k pk |ψk ⟩⟨ψk |{i,j} is a symmetric state,
then R(|ψk ⟩⟨ψk |{i,j} ) ⊆ R(ρ{i,j} ). We obtain that |ψk ⟩⟨ψk |{i,j} is also a symmetric state for each k. From the above
discussion, Swap(i,j) |ψk ⟩ = |ψk ⟩ for each k, which means that Swap(i,j) ρ = ρ.
If ρS is a symmetric state, then ρ{s,t} is also a symmetric state for each {s, t} ⊆ S by Lemma 1. By the above
discussion, Swap(s,t) ρ = ρ for each {s, t} ⊆ S. If the hypergraph G = ([n], S) is connected, then for each {a1 , am } ⊆
[n], there exists a path
a1 ∼ a2 ∼ a3 ∼ · · · am (20)
such that {aj , aj+1 } is contained in some S for 1 ≤ j ≤ m − 1, where S ∈ S. We obtain that Swap(aj ,aj+1 ) ρ = ρ for
1 ≤ j ≤ m − 1. Since
Swap(a1 ,a3 ) = Swap(a1 ,a2 ) Swap(a2 ,a3 ) Swap(a1 ,a2 ) , (21)
we have Swap(a1 ,a3 ) ρ = ρ. Next, since
Swap(a1 ,a4 ) = Swap(a1 ,a3 ) Swap(a3 ,a4 ) Swap(a1 ,a3 ) , (22)
we also have Swap(a1 ,a4 ) ρ = ρ. By repeating this process m − 1 times, we obtain Swap(a1 ,am ) ρ = ρ. Thus
Swap(a1 ,am ) ρ = ρ for each {a1 , am } ⊆ [n], i.e. ρ is a symmetric state.

Appendix C: The proof of Theorem 1

Theorem 1 For all 1 ≤ k ≤ n − 1 and n ≥ 3, S detects the Dicke state |Dnk ⟩’s GME if and only if the hypergraph
G = ([n], S) is connected. Moreover, l(|Dnk ⟩) = 2, and m(|Dnk ⟩) = n − 1.
Proof. Since the necessity is obtained from Lemma 1 in the main text, we only need to show the sufficiency. Let
S be a collection of subsets, with S ⊆ [n], ∀S ∈ S. Assume G = ([n], S) is connected and ρ ∈ C(|Dnk ⟩, S), we need to
show that ρ is a genuinely entangled state. By Lemma 1, ρS = |Dnk ⟩⟨Dnk |S is a symmetric state for all S ∈ S. Since
the hypergraph G = ([n], S) is connected, ρ is also a symmetric state by Lemma 2. We can write ρ as
n
X
ρ= ai,j |Dni ⟩⟨Dnj |, (23)
i,j=0
 9

where A = (ai,j )0≤i,j≤n is positive and Tr(A) = 1.

Since there exists an S ∈ S with |S| ≥ 2 and {a, b} ⊆ S, we have ρ{1,2} = ρ{a,b} = |Dnk ⟩⟨Dnk |{a,b} = |Dnk ⟩⟨Dnk |{1,2}
Pn
by Lemma 1. Then we obtain i,j=0 ai,j |Dni ⟩⟨Dnj |{1,2} = |Dnk ⟩⟨Dnk |{1,2} , and this implies that
n
X
|Dnk ⟩⟨Dnk |{1,2} = ai,i |Dni ⟩⟨Dni |{1,2} . (24)
i=0

There are five cases for |Dni ⟩⟨Dni |{1,2} .

(i) When i = 0,
|Dn0 ⟩⟨Dn0 |{1,2} = |D20 ⟩⟨D20 |. (25)

(ii) When i = 1,
n−2 0 2
|Dn1 ⟩⟨Dn1 |{1,2} = |D2 ⟩⟨D20 | + |D21 ⟩⟨D21 |. (26)
n n
(iii) When 2 ≤ i ≤ n − 2,
n−2 n−2 n−2




2 i−1 i−2
|Dni ⟩⟨Dni |{1,2} = i

n |D20 ⟩⟨D20 | + n

|D21 ⟩⟨D21 | + n

|D22 ⟩⟨D22 |. (27)
i i i

(iv) When i = n − 1,
2 1 n−2 2
|Dnn−1 ⟩⟨Dnn−1 |{1,2} = |D ⟩⟨D21 | + |D2 ⟩⟨D22 |. (28)
n 2 n
(v) When i = n,
|Dnn ⟩⟨Dnn |{1,2} = |D22 ⟩⟨D22 |. (29)

Then
n n−2 n−2

! n−1 n−2

! n n−2

!
X X X 2 i−1
X
ai,i |Dni ⟩⟨Dni |{1,2} = ai,i ni
|D20 ⟩⟨D20 | + ai,i n

|D21 ⟩⟨D21 | + ai,i i−2
n

|D22 ⟩⟨D22 |, (30)
i=0 i=0 i i=1 i i=2 i

There are three cases for Eq. (24).

(1) If k = 1 and n ≥ 3, then we obtain that
n−2 n−2 n−1 n−2 n n−2




X n−2 X 2 i−1 2 X
ai,i ni
= , ai,i n

= , ai,i i−2
n

= 0. (31)
i=0 i
n i=1 i
n i=2 i

Since A = (ai,j )0≤i,j≤n is positive, it must have ai,i ≥ 0 for all 0 ≤ i ≤ n. By Eq. (31), we obtain that a1,1 = 1,
and ai,i = 0 for all i ̸= 1. Since A is positive, it must have ai,j = aj,i = 0 for all 0 ≤ i ̸= 1 ≤ n, and 0 ≤ j ≤ n.
Thus ρ = |Dn1 ⟩⟨Dn1 |.
(2) If 2 ≤ k ≤ n − 2, and n ≥ 4, then we obtain that
n−2 n−2 n−2 n−1 n−2 n−2 n n−2 n−2







X X 2 i−1 2 k−1
X
i

ai,i n = n , k

ai,i n

= n

, ai,i i−2
n

= k−2
n

. (32)
i=0 i k i=1 i k i=2 i k

These three equations can be simplified into the following three equations,
n
X
ai,i (n − i)(n − i − 1) = (n − k)(n − k − 1), (33)
i=0
Xn
ai,i i(n − i) = k(n − k), (34)
i=0
Xn
ai,i i(i − 1) = k(k − 1). (35)
i=0
 10

Next, by adding both sides of Eq. (33) and Eq. (34), and dividing both sides by n−1 (i.e. (Eq. (33) + Eq. (34))/(n−
1)). we obtain the following equation,
n
X
ai,i (n − i) = (n − k). (36)
i=0

The equation (Eq. (34) + Eq. (35))/(n − 1) is

n
X
ai,i i = k. (37)
i=0

The equation Eq. (33) + Eq. (36) is

n
X
ai,i (n − i)2 = (n − k)2 . (38)
i=0

The equation Eq. (35) + Eq. (37) is

n
X
ai,i i2 = k 2 . (39)
i=0
Then
n
! n
! n
!2
X X X
2 2 2 2
k (n − k) = ai,i i ai,i (n − i) ≥ ai,i i(n − i) = k 2 (n − k)2 (40)
i=0 i=0 i=0

where the inequality is the Cauchy–Schwarz inequality, and the last equation is obtained from Eq. (34). Since the
Cauchy–Schwarz inequality has the“=” sign, there must exist a real number s ̸= 0, such that
√ √
ai,i i = s ai,i (n − i), ∀0 ≤ i ≤ n. (41)
Then a0,0 = an,n = 0, and there exists only one nonzero aj,j P for all 1 ≤ j ≤ n − 1 (i.e. aj,j ̸= 0, and ai,i = 0 for
n
all 1 ≤ i ̸= j ≤ n − 1). Moreover, since Tr(A) = 1, we have i=0 ai,i = 1. It implies that aj,j = 1, and ai,i = 0
for all 0 ≤ i ̸= j ≤ n. By using Eq. (37), we obtain that j = k, that is ak,k = 1, and ai,i = 0 for all 0 ≤ i ̸= k ≤ n.
Since A is positive, it must have ai,j = aj,i = 0 for all 0 ≤ i ̸= k ≤ n, and 0 ≤ j ≤ n. Thus ρ = |Dnk ⟩⟨Dnk |.
(3) If k = n − 1 and n ≥ 3, then we obtain that
n−2 n−2 n−1 n−2 n n−2




X X 2 i−1 2 X n−2
i

ai,i n = 0, ai,i n

= , ai,i i−2
n

= . (42)
i=0 i i=1 i
n i=2 i
n
It implies that an−1,n−1 = 1, and ai,i = 0 for all i ̸= 1. Since A is positive, it must have ai,j = aj,i = 0 for all
0 ≤ i ̸= n − 1 ≤ n, and 0 ≤ j ≤ n. Thus ρ = |Dnn−1 ⟩⟨Dnn−1 |.
Above all, the compatibility set C(|Dnk ⟩, S) contains only |Dnk ⟩, which is a genuinely entangled state. Therefore, S
detects |Dnk ⟩’s GME. By Eq. (4) and Eq. (5) in the main text, we can easily obtain that l(|Dnk ⟩) = 2 and m(|Dnk ⟩) =
n − 1.

Appendix D: State determination length

Definition 2 Given an n-qubit ρ, and a collection of subsets S, with S ⊆ [n], ∀S ∈ S, if the compatibility set C(ρ, S)
contains only ρ, then we say that S uniquely determines ρ.
We denote l(S) := maxS∈S |S| as the maximum size of a subset in S. The state determination length of the state ρ is
then defined as

L(ρ) := min l(S) S uniquely determines ρ . (43)
S
We also denote

M (ρ) := min |S| S uniquely determines ρ . (44)
S: |S|=L(ρ) ,∀S∈S

as the minimum number of L(ρ)-body marginals needed to uniquely determine ρ.

 11

Appendix E: The proof of Theorem 2

Theorem 2 Let

|ψ⟩ = a|Wn ⟩ + b|11 · · · 1⟩, (45)

where ab ̸= 0, |a|2 + |b|2 = 1, and n ≥ 4. Then L(|ψ⟩) = n − 1, M (|ψ⟩) = 2. Moreover,

(i) If |ψ⟩⟨ψ|{1,2} is NPT, then

l(|ψ⟩) = 2, m(|ψ⟩) = n − 1, gap(|ψ⟩) = n − 3.

(ii) If |ψ⟩⟨ψ|{1,2} is PPT, then



n−1
l(|ψ⟩) = 3, m(|ψ⟩) = , gap(|ψ⟩) = n − 4.
2

Proof. Firstly, we compute L(|ψ⟩) and M (|ψ⟩). Obviously, |ψ⟩ is not a GHZ-like state. By previous discussion,
we have L(|ψ⟩) ≤ n − 1. However, the compatibility set C(|Dnk ⟩, S1 ) (S1 consists of all (n − 2)-subsets of [n]) contains
the mixed state

σ = |a|2 |Wn ⟩⟨Wn | + |b|2 |11 · · · 1⟩⟨11 · · · 1|.

Since rank(σ) = 2, we have σ ̸= |ψ⟩⟨ψ|. Then we obtain that S1 cannot uniquely determine |ψ⟩. Thus L(|ψ⟩) = n − 1.
From the proof of Lemma 1 in the main text, we know that if S can uniquely determine an n-qubit genuinely
entangled state, then the hypergraph G = ([n], S) must be connected. This means M (|ψ⟩) ≥ 2. Assume S2 consists
of (n − 1)-subsets of [n] with |S2 | ≥ 2 and ρ ∈ C(|ψ⟩, S2 ). Since |ψ⟩ is a symmetric pure state, ρS = |ψ⟩⟨ψ|S is also
a symmetric state for all S ∈ S2 by Lemma 1. By Lemma 2, we know that ρ is a symmetric state. It implies that
ρ{1,2,...,n−1} = ρS = |ψ⟩⟨ψ|S = |ψ⟩⟨ψ|{1,2,...,n−1} for all S ∈ S2 by Lemma 1, which means that there is only one
relation ρ{1,2,...,n−1} = |ψ⟩⟨ψ|{1,2,...,n−1} for any |S2 | ≥ 2. Since L(|ψ⟩) = n − 1 and 2 ≤ |S2 | ≤ n, we have M (|ψ⟩) = 2.
Next, we compute l(|ψ⟩) and m(|ψ⟩). Assume S3 consists of 2-subsets of [n] (the corresponding hypergraph G =
([n], S3 ) is connected), and ρ ∈ C(|ψ⟩, S3 ). By Lemmas 1 and 2, we know that ρ is also a symmetric state. Thus ρ is
either a genuinely entangled state or a fully separable state. According to Lemma 1, ρS = ρ{1,2} , ∀S ∈ S3 .
There are two cases.
(i) If ρ{1,2} = |ψ⟩⟨ψ|{1,2} is NPT, then ρ{1,2} is an entangled state. This implies that ρ must be a genuinely entangled
state, and C(|ψ⟩, S3 ) contains only genuinely entangled states. Thus l(|ψ⟩) = 2, and m(|ψ⟩) = n − 1.
If ρ{1,2} = |ψ⟩⟨ψ|{1,2} is PPT, then ρ{1,2} is a separable state, which can be written as ρ{1,2} =
(ii) P
i |αi αi ⟩⟨αi αi |. Let S4 be the set of all 2-subsets of [n], then C(|ψ⟩, S4 ) contains a fully separable state
i pP
τ = i pi |αi αi · · · αi ⟩⟨αi αi · · · αi |. This implies that l(|ψ⟩) ≥ 3. We assume S5 consists of 3-subsets of [n] (the
corresponding hypergraph G = ([n], S5 ) is connected), and ρ ∈ C(|ψ⟩, S5 ). By the similar discussion as above,
we known that ρ is either a genuinely entangled state or a fully separable state. Consider

3|a|2 (n − 3)|a|2
ρ{1,2,3} = |ψ⟩⟨ψ|{1,2,3} = |W3 ⟩⟨W3 | + |000⟩⟨000| + |b|2 |111⟩⟨111|.
n n
Note that R(ρ{1,2,3} ) is spanned by {|W3 ⟩, |000⟩, |111⟩}, and Dim(R(ρ{1,2,3} )) = 3. Since all biproduct states
in R(ρ{1,2,3} ) are {a|000⟩, b|111⟩ : |a| = |b| = 1} and they cannot span R(ρ{1,2,3} ), we obtain that ρ{1,2,3} is a
genuinely entangled state. Then ρ must be a genuinely  entangled state, and C(|ψ⟩, S5 ) contains only genuinely
entangled states. Thus l(|ψ⟩) = 3 and m(|ψ⟩) = n−1

2 .