Multipartite entanglement structures in quantum stabilizer states

Vaibhav Sharma∗ and Erich J Mueller†

Laboratory of Atomic and Solid State Physics, Cornell University, Ithaca, New York
(Dated: November 6, 2024)
We develop a method for visualizing the internal structure of multipartite entanglement in pure
stabilizer states. Our algorithm graphically organizes the many-body correlations in a hierarchical
structure. This provides a rich taxonomy from which one can extract a number of traditional quan-
tities such as entanglement depth and entanglement entropy. Our construction is gauge invariant
and goes beyond traditional entanglement measures by visually revealing how quantum information
and entanglement is distributed. We use this tool to analyze the internal structures of prototypical
stabilizer states (GHZ state, cluster state, stabilizer error correction codes) and are able to contrast
the complexity of highly entangled volume law states generated by random unitary operators and
random projective measurements.
arXiv:2411.02630v1 [quant-ph] 4 Nov 2024

I. INTRODUCTION of least (w −1) other qubits within the same cluster. The
idea is that w qubits together encode information that
Entanglement is the key feature that distinguishes is not present in any individual qubit. We then recur-
quantum and classical systems. It is a valuable resource sively continue to group these clusters together, forming
for quantum information processing and computation [1]. an entanglement structure diagram that shows how var-
When restricted to two parties, entanglement is well un- ious parts of the system are connected. There is some
derstood: Bipartite entanglement measures are very intu- relationship between this concept and the idea of the in-
itive, and there is no controversy about how to detect or formation lattice, which additionally incorporates spatial
quantify two-particle entanglement. For many-body sys- information [11–13].
tems, the situation is different. Characterizing and un- Following this approach, we are able to directly vi-
derstanding multipartite entanglement is challenging due sualize how quantum information is spread throughout
to its far richer structure that cannot be easily summa- the system in highly-entangled states. We get access to
rized [2]. Here we develop a more intuitive and broadly the internal structure of correlations that connect vari-
applicable approach towards organizing and visualizing ous qubits in a composite many-body state. We note that
multipartite entanglement in stabilizer states. the recursive nature of our method makes extracting such
There exist several measures of multipartite entangle- information much more efficient than simply calculating
ment such as the entanglement entropy [3], entanglement bipartite entanglement entropy across all possible parti-
depth [4, 5], global entanglement [6], quantum Fisher in- tions of a given system. Our method also provides the
formation [7], Schmidt measure [8], generalized geomet- entanglement depth and can be used to bound entangle-
ric measure [9] and N-tangle [10]. Each of these mea- ment entropies.
sures emphasize a different aspect of multipartite entan- We illustrate our decomposition by looking at well-
glement, quantifying it with a single number. Although known stabilizer states: the cluster state, GHZ state
they provide useful information, they fail to fully capture and the logical states of error correction codes such as
the complexity of a multipartite entangled state, such as the 5-qubit code and the 7-qubit CSS code. The true
the internal structure of correlations and the local distri- power of our construction, however, is revealed by look-
bution of entanglement. This is especially apparent for ing at highly entangled volume law states generated by
states that are not described by traditional order param- random two-qubit Clifford unitaries and random three-
eters, such as those generated by noisy quantum circuits. qubit projective measurements [14, 15]. Although both
In this work, our goal is to resolve the structure of an sets of states show volume law scaling of bipartite entan-
entangled state and develop a way to visualize multipar- glement entropy, we find that the entanglement structure
tite entanglement, going beyond reducing it to a single diagrams are able to distinguish between them.
numerical quantity. We propose organizing multipartite
entanglement via the topological structure of correlations
present in the state. We group the qubits in a quantum II. ALGEBRAIC STRUCTURE
state in clusters characterized by a parameter, w. The
defining features of these w-clusters is that each qubit in- Our approach works for stabilizer states [16], which are
side them has non-zero mutual information with a subset an important class of quantum states which are used in
quantum computing and play an important role in quan-
tum error correction. An N qubit stabilizer state is the
simultaneous eigenstate of N linearly independent com-
∗ vaibhavsharma@[Link]; Current Address: Department of muting Pauli strings, referred to as the
√ stabilizer gener-
Physics, Rice University, Houston, Texas ators. For example, the cat state, 1/ 2(|111⟩ + |000⟩) is
† em256@[Link] the simultaneous eigenstate of three Pauli string opera-
 2

tors, Z1 Z2 I3 , I1 Z2 Z3 and X1 X2 X3 . Trivially, a stabilizer be disjoint from all other such w-element subsets within
state is also an eigenstate of any member of the stabilizer A. Equivalently, every element of A is in the support of
group formed by taking arbitrary products of the gener- an A-stabilizer of weight less than or equal to w. The
ators. The stabilizer generators are not unique, and can region A is a w-cluster if it is both w-indivisible and w-
be replaced by any N linearly independent elements of connected.
the stabilizer group. This freedom to choose the gener- In our construction, we use an iterative method to di-
ators is often referred to as a gauge freedom. We only vide the system into w-clusters. In the first iteration,
consider pure quantum states, and will not discuss the each qubit is considered as a single element. At each
properties of ensembles encoded by density matrices. subsequent iteration, any w-clusters found in the previ-
A generic Pauli string can be written as P = ous iteration are treated as single indivisible elements.
Q αj βj
j Xj Zj , where αj , βj = 0, 1. The support of a Pauli We begin by finding all of the qubits which are disentan-
string is the set of qubits for which α and/or β is non- gled from the rest of the system. These are 1-clusters.
zero. The weight of the string is the number of qubits These singletons are decoupled from the rest of the sys-
in the support. The expectation value of a weight-w sta- tem and can be ignored at future steps. Next we find all
bilizer can be interpreted as the expectation value of a of the qubits which belong to 2-clusters. Formally these
w-spin correlation function. In our approach, we recur- can be found by constructing all stabilizers of weight 2,
sively group sets of qubits into clusters. At each iteration, or by looking at the reduced density matrix for each pair.
we take the weight of a stabilizer to be the number of The sets of elements which are connected by these sta-
clusters in its support, rather than the number of qubits. bilizers form the 2-clusters. At this point we treat the
A key feature of stabilizer states is that the bipartite clusters as indivisible elements. We then repeat the pre-
entanglement entropy is quantized. If we break a stabi- vious steps – removing single elements which are decou-
lizer state into two disjoint sets of qubits, the entangle- pled, and forming new 2-clusters of the given elements.
ment entropy across the cut is always a multiple of ln(2). We iterate until no new 2-clusters can be formed. We
If region A contains nA qubits then SA ≤ nA ln(2). When then search for 3-clusters, if there are any elements which
this bound is saturated, we refer to A as being maximally have not yet been assigned to decoupled clusters. If any
entangled. One learns nothing about the state of the sys- 3-clusters are found, we repeat all the previous steps. We
tem by interrogating region A. There is no information always construct as many low weight clusters as possible
which is exclusively stored in A, and the reduced density before moving on to higher weight clusters. This proce-
matrix is the identity matrix of dimension 2nA . dure can be repeated with 4-clusters, 5-clusters... until
A useful intuition is that if SA = sa ln(2) with respect all qubits have been assigned to large decoupled clusters.
to the rest of the system, then there are nA −sa bits of in- We give more details of the computational procedure in
formation that are locally stored within A. In particular, Appendix A.
it is an eigenstate of exactly nA − sa linearly independent A short example is useful. Consider the 4-qubit state,
Pauli strings whose support is entirely within A. These |ψ⟩ with stabilizer generators Z1 Z2 , Z3 Z4 , X1 X2 Z3 and
are the operators which measure the stored information. Z2 X3 X4 . Up to a normalization constant, |ψ⟩ = |1111⟩+
They generate a subgroup of the stabilizer group, which |0011⟩ + |1100⟩ − |0000⟩. The entanglement structure of
we refer to as the A-stabilizers. Another useful feature is this state is shown in Fig. 1. Here the qubits are labeled
that the reduced density matrix of A has rank 2sa . by black integers. The w-clusters are denoted by ovals,
We can generalize this notion to the case when region A with w shown in red. In the first round, our algorithm
is divided into disjoint clusters of qubits, constructed by produces two 2-clusters: (1, 2) and (3, 4). In the second
some as-yet unspecified algorithm. Each cluster will be round, we consider the clusters (1, 2) and (3, 4) as sin-
labeled by an index i. Following WatanabeP[17], we
define gle indivisible elements. We then see that these together
the total correlations in A to be IA = i∈A Si − SA
form one large 2-cluster: ((1, 2), (3, 4)) connecting all four
where Si is the entanglement entropy of cluster i with qubits into a single many-body state. Such a structure
the rest of the system. The total correlations are also implies that while qubits 1, 2 and 3, 4 are directly con-
referred to as the “multipartite quantum mutual infor- nected by weight-2 stabilizers, qubits 1, 3 and qubits 1, 4
mation” [18]. It tells us how many bits of information are only connected by higher weight stabilizers. In this
are stored in A, but which are not stored solely by any way our construction recursively forms clusters to build
single cluster within A. a complete description of entanglement within the many-
We now coin three new terms: w-indivisible, w- body state.
connected and w-cluster. We say that region A is w- Fig. 2 shows a more complicated entanglement struc-
indivisible if any subset C ⊂ A containing w − 1 or ture diagram of a 16-qubit state. In this diagram, qubit
fewer elements have vanishing total correlations, IC = 0. 16 forms an isolated cluster. This means that it is not
Equivalently, any stabilizers whose support lies solely in entangled with any other qubit. Qubits labeled 13, 14
A must have a weight greater or equal to w. and 15 similarly form an independent cluster which is
We say that A is w-connected if every element j ∈ A be- unentangled with the rest of the system. It is labeled
longs to a set B ⊆ A containing w elements, such that B as a 2-cluster, indicating that each of these qubits must
has non-zero total correlations. Furthermore, B cannot be in the support of a weight 2 stabilizer, whose support
 3

m regions, the state is deemed to have an entanglement

depth of k [4, 5]. The state is said to have k-qubit mul-
1 2 3 4
2 2 tiparticle entanglement. In our entanglement structure
2 diagram, the entanglement depth is simply the number
of qubits in the largest cluster.
In the four qubit example shown in Fig. 1, all the qubits
FIG. 1. Entanglement structure diagram of a four qubit state are ultimately part of one big cluster, denoting an entan-
given by the wavefunction, |ψ⟩ = 1/2(|1111⟩+|0011⟩+|1100⟩− glement depth of 4. In contrast, the diagram in Fig. 2
|0000⟩). The qubits, labeled by black integers, are placed in shows that the 16-qubit state can be written as a product
clusters denoted by drawing circles around them. The clus- state of the form, |ψ⟩ = |ϕ1 ⟩A1 ⊗ |ϕ2 ⟩A2 ⊗ |ϕ3 ⟩A3 where
ters are then considered indivisible entities and can be fur-
region A1 contains qubits labeled from 1-12, A2 contains
ther placed into bigger clusters. The small red number, w in
the bottom left of a clustering circle denotes that the entities
qubits 13-15 and A3 contains the qubit 16. Clearly this
inside that cluster share a minimum of w-point correlation state has an entanglement depth of 12.
function among the entities.

B. Minimal stabilizer weight

is fully within this cluster. Up to local Pauli rotations,
an unentangled 2-cluster of three spins is a GHZ state
|111⟩ + |000⟩. Another important feature that can be extracted from
The middle cluster containing qubits from 1-12 is the these diagrams is the weight of the smallest stabilizer in
largest, with multiple sub-clusters. At the first level, the stabilizer group. We refer to this as the minimal sta-
qubits (1,4), qubits (2,3) and qubits (5,12) are placed bilizer weight. It reveals the degree of delocalizion of the
in 2-clusters, denoting presence of weight 2 stabilizers. quantum information. In the entanglement diagram, one
By virtue of being 2-clusters that are not isolated, each can identify the minimal stabilizer weight as the smallest
of them contains one bit of information and has an en- w, for the w-clusters in the first iteration of our algo-
tanglement entropy S = ln 2 with the rest of the system. rithm.
In the next iteration, the 2-clusters formed by qubits
(1,4) and qubits (2,3) join together to form a bigger 2-
cluster. There is a stabilizer whose support is in both of C. Bipartite entanglement entropy
these clusters but that cannot be individually confined
to either one of these. Similarly, the 2-cluster of qubits We can also use the diagram to obtain an upper bound
(5,12) joins with qubit 6 to form a bigger 2-cluster. on the bipartite entanglement entropy across any given
We can further deduce constraints from the fact that partition. Consider a set of n qubits that are partitioned
(1, 4) and (2, 3) are 2-clusters embedded in a larger 2- into two regions, region A containing m and region B
cluster. There is a stabilizer whose support lies in containing n − m qubits with m < n − m. The maximum
(1, 2, 3, 4). It must have support on at least 3 of these entanglement entropy across the partition in a stabilizer
qubits – as otherwise (1, 2, 3, 4) would have all been state is m ln 2 since there are only m units of information
placed in one 2-cluster during the first iteration. available in region A that can be shared with the other
In the next iteration, the clusters of qubits ((1,4),(2,3)) region B. Within region A, any local clusters consisting
and qubits (5,6,12) further join to form a new 2-cluster. of qubits entirely within the region reduce the entangle-
At the final level, this newly formed cluster joins with ment entropy across the partition by at-least one unit.
the 5 individual qubits labeled from 7-11 in a 3-cluster This is because such a local cluster denotes a stabilizer
to form a big independent cluster containing qubits 1- that is completely confined within the region A. Identi-
12. This is labeled a 3-cluster, meaning that any of the fying such local clusters gives upper bounds to the en-
elements can be joined up with two others to form a set tanglement entropy of a given region with the rest of the
which contains the support of a stabilizer. system.
The four qubit example of Fig. 1 is the easiest to un-
A. Entanglement depth
derstand. Consider region A with qubits (1, 2) and region
B with qubits (3, 4). Simply by counting the number of
qubits we know that the maximum possible bipartite en-
Having constructed an entanglement diagram, we can tanglement entropy between A and B is 2 ln 2. Beyond
readily extract the entanglement depth [4, 5]. Any n- that, within region A there is one local cluster containing
qubit pure state can be written as a product state over one unit of information entirely within A. This reduces
m disjoint regions in the following form, the entanglement entropy across the partition by ln 2.
|ψ⟩ = |ϕ1 ⟩A1 ⊗ |ϕ2 ⟩A2 ..... ⊗ |ϕm ⟩Am (1) Thus we know that the entanglement entropy across the
partition is bounded from above by ln 2. In this case the
where none of the |ϕj ⟩Aj can be further decomposed into bound is saturated, as the two subclusters belong to one
product states. If there are at most k qubits in any of the bigger cluster, and hence share one bit of information.
 4

8
7 9

1 4 2 3 5 12 6 13 14 15
16 2 2 2 2 2
2 2

3
10 11

FIG. 2. An entanglement structure diagram of an arbitrary 16 qubit stabilizer state is shown. The state is separable with
qubits labeled 1-12 forming the largest cluster, corresponding to an entanglement depth of 12. The nested structure of the
clusters shows how correlations are distributed in the state.

Now consider the example shown in Fig. 2, where

qubits 1-12 are in a pure state. We can separate that
pure state into two regions – for example, we take region (a) 1 2 3 4
A to contain qubits 1-6 and region B to contain qubits
7-12. From counting qubits, the maximum possible en- 5 6 7 8
tanglement entropy across this partition is 6 ln 2. Within 2
region A, we see two sub-clusters that are 2-clusters con-
taining qubits (1,4) and qubits (2,3). They reduce the
entanglement entropy across the partition by 2 ln 2. They
are both further joined in another local 2-cluster, reduc-
(b)
ing the entropy further by ln 2. Thus the upper bound on
entanglement entropy between regions A and B is 3 ln 2. 1 2 3 4
2 2 2
III. ENTANGLEMENT STRUCTURES OF
SOME PROTOTYPICAL STATES
5 6 7 8
2 2 2
We can use the entanglement structure diagrams to vi- 2
sualize some prototypical stabilizer states.
√ The diagram
for a GHZ state [19] of 8 qubits, |ψ⟩ = 1/ 2(|11111111⟩+
|00000000⟩) is shown in Fig. 3(a). Among the eight sta-
bilizer generators of the GHZ state, seven can be taken
to have the form, Si = Zi Zi+1 where i ⊂ [1, ., 7]. The FIG. 3. Entanglement structure diagram of (a) 8-qubit GHZ
eighth generator can be chosen to be S8 = X1 X2 ...X8 . state and (b) 8-qubit cluster state in 1D without periodic
boundary conditions
Here Zi , Xi are Pauli operators σiz , σix for the qubit la-
beled i. All the qubits share one big 2-cluster, denoting
the presence of 2-point correlations (weight-2 stabilizers)
and a full entanglement depth of 8. In fact, in a GHZ Both the GHZ and the cluster state are low entan-
state, each qubit pair (i,j) is in the support of a weight glement area law states. We know that across a bipar-
two stabilizer, Zi Zj . tition in the center, they only have ln(2) entanglement
In Fig. 3(b) we show the entanglement structure dia- entropy yet the entanglement diagrams have very differ-
gram for the 8-qubit cluster state on a 1D lattice without ent structures. The ln(2) entanglement of the cluster
periodic boundary conditions [20]. This is defined by its state can be extracted from the reasoning in Sec. II C.
stabilizer generators: Sj = Zj−1 Xj Zj+1 for j ̸= 1, 8 and We imagine a bipartition into the regions A = (1, 2, 3, 4)
S1 = X1 Z2 , S8 = Z7 X8 . These latter 2 generators have and B = (5, 6, 7, 8). We can iteratively find the total
weight 2, and hence both (1, 2) and (7, 8) are placed in correlations of region A: There is one bit of information
2-clusters. Treating these 2-clusters as indivisible, the re- stored in (1, 2). Every time we grow that cluster, we add
maining stabilizers have the same topological structure. one qubit, but also add one confined stabilizer. Hence
Thus ((1, 2), 3) and (6, (7, 8)) form 2 clusters. We can re- ((1, 2), 3) and (((1, 2), 3), 4) each contain one bit of infor-
peat, until one large cluster is formed. The local nature mation. Consequently the bipartition into A and B has
of the correlations is apparent in the nested clusters. As an entanglement entropy of only ln(2). The entanglement
with the GHZ state, the entanglement depth is 8. entropy of the GHZ state, however, cannot be extracted
 5

(a) 1 2 3 (a) 7 1 2 3 5 6 8 11
4 14 20
4 5 3 10 9 12 13 15 18 19 5 16
2

(b) 1 2 3 4 (b)
5 6 7 3
14
10 1 2 3 4 5 6 7 8 9
2 2 3 2

FIG. 4. Entanglement structure diagram of (a) the five-qubit

error correcting code and (b) seven-qubit Steane CSS error 15 16 17 2
correcting code 2 2
2
18 19 20 11 12 13
3 3
solely from the entanglement structure diagram.
Fig. 4(a),(b) show the entanglement structure dia-
grams of the five-qubit and seven-qubit error correcting FIG. 5. Entanglement structure diagram of a typical 20 qubit
codes. The stabilizer generators are given in Ref. [16] volume law state generated by (a) local two-qubit random
and these codes can correct arbitrary single qubit errors. Clifford unitary gates and (b) local three-qubit random pro-
The figures show the diagrams in the case when the log- jective measurements
ical qubit is in the eigenstate of the logical X operator.
In both cases, all the qubits are in one 3-cluster showing 𝑺𝑬𝑬
an entanglement depth of 5 and 7 respectively. We can
conclude that each of the qubits is in support of a weight
3 stabilizer that can be used to interrogate whether the Clifford Unitary
system is in the code space. No extra information is Three site measurements
given by the entanglement diagram, and one can argue
that these diagrams have limited utility for such a class
of states.

IV. ENTANGLEMENT STRUCTURES OF

HIGHLY ENTANGLED VOLUME LAW STATES
L
We now analyze the internal structure of highly en-
tangled volume law states generated in random quan-
FIG. 6. Average bipartite entanglement entropy, SEE as a
tum circuits. Random quantum circuits have recently
function of system size L for states generated by two-qubit
been instrumental in enhancing our understanding of random Clifford unitary operators (blue) and three-qubit ran-
generic quantum dynamics and novel entanglement phase dom measurements (red). SEE scales linearly with L in both
transitions [21]. In 1D systems, an entanglement phase cases.
transition between area law and volume law phases oc-
curs when single site projective measurements are inter-
spersed with random two-qubit unitary operators [14, 22– measurements to be random weight-3 Pauli strings act-
24]. Furthermore, there have been various measurement- ing on neighboring sites. In our analysis, we start from
only phase transitions from volume law to area law [15] a product state and apply our random unitary gates or
or between different area law phases using random non- measurements until a steady state distribution is pro-
commuting projective measurements [25–31]. The states duced. We generate 500 states in both categories for var-
generated by typical random quantum circuits do not ious systems sizes, L. In Fig. 6, we plot the average bi-
have an order parameter and are characterized by the partite entanglement entropy, SEE between two halves of
scaling of their bipartite entanglement entropy. Not the system as a function of system size L. In both cases,
much is known about their spatial structure or proper- SEE ∝ L, signalling a volume law. For the same sys-
ties. The entanglement structure diagrams can fill this tem sizes, SEE is greater for the unitarily evolved state.
gap. In fact, the states generated by unitary evolution typi-
We focus on volume law states in 1D generated by ran- cally saturate SEE such that, SEE ∼ L/2. These are
dom local two-qubit Clifford unitary gates or random lo- referred to as Page states [32]. This maximal entangle-
cal three-qubit projective measurements. We take these ment entropy reflects the behavior of a quantum state
 6

(a) Depth weight makes it expensive to construct entanglement di-

agrams for the unitary volume law states. Thus we are
limited to moderate system sizes, L ∼ 24 for unitary
volume law states due to this increasing complexity.
The entanglement structure diagram enables us to
visualize the differences between the entanglement en-
tropy of the unitary and measurement only volume law
states. Fig. 5 shows entanglement structure diagrams
Clifford Unitary for a generic unitary and measurement-only volume law
state of 20 qubits. The measurement-only state has local
Three site measurements
sub-clusters that do not have correlations going across
the boundary at the middle. These local sub-clusters
L
(b) n dilute the long-range spreading of quantum information
and reduce the bipartite entanglement entropy. The uni-
tary volume law states have no such local sub-clusters,
Clifford Unitary thus allowing full scrambling of information and maximal
entanglement entropy. While both random unitaries and
Three site measurements random measurements can produce volume law states,
there is more localized information present in the case of
measurements.

V. SUMMARY AND OUTLOOK

L
It is challenging to understand and visualize multipar-
FIG. 7. (a) Average entanglement depth and (b) minimal sta- tite entanglement, and the literature contains a number
bilizer weight n of volume law states generated by two-qubit of entanglement measures, each of which are useful in ap-
random Clifford unitary operators (blue) and three-qubit ran- propriate circumstances [2]. In this paper we introduced
dom measurements (red). Both kinds of volume law states a method for characterizing the multipartite entangle-
have similar entanglement depth. However the minimal sta- ment of stabilizer states. It allows us to visualize how
bilizer weight found in the entanglement structure diagram
quantum information and entanglement is internally dis-
scales linearly for unitarily evolved states while it remains
constant for measurement-only states tributed among the qubits of a many-body state.
The most immediate application of our technique is
understanding the volume law states which are produced
in random quantum circuits. None of the traditional en-
under chaotic evolution where the quantum information tanglement measures are particularly insightful for un-
is maximally scrambled. In contrast, the states gener- derstanding the properties of these states. Our approach
ated purely by measurements have a lower entanglement quantifies the way in which information can be extracted
entropy, but still follow the volume law. We will see how from them when clusters of qubits are interrogated. The
the entanglement structure diagrams can help us to vi- size and distribution of these clusters helps us identify
sualize this difference. key differences between volume law states generated by
We can use the entanglement structure diagrams unitary operators vs measurements.
to probe the internal structures of unitary and Rather than using a single number to characterize the
measurement-only volume law states as a function of sys- entanglement, we produce a diagram, which groups the
tem size, L. Fig. 7(a) shows the average entanglement qubits into a hierarchical arrangements of clusters. Each
depth and Fig. 7(b) shows the minimal stabilizer weight, cluster is labeled by an integer which specifies the weight
n (see Sec. II B). The average is taken over all 500 states of the smallest stabilizer connecting the elements con-
for each system size. Although both categories of states tained in it. From these diagrams we can extract the en-
show a full entanglement depth, ∼ L, they differ sharply tanglement depth, and bound the entanglement entropy
in their minimal stabilizer weight. This weight grows lin- across an arbitrary cut. We can also read off the ’mini-
early with system size for volume law states generated by mal stabilizer weight,’ which is the weight of the smallest
unitary operators while it stays constant for states gener- stabilizer.
ated by measurements only. In the states generated by lo- The entanglement structure diagrams are trivial for
cal measurements, the state always contains 2-clusters or many classic stabilizer states. Most states appearing in
3-clusters. The information stored in the unitary volume quantum error correcting codes will have an entangle-
law states is more delocalized – extracting any informa- ment structure diagram containing only a single cluster.
tion requires measuring an extensive number of qubits. The GHZ state also has this structure. The 1D cluster
As is detailed in Appendix B, the large minimal stabilizer state with hard wall boundaries, however, has an entan-
 7

glement structure diagram which reflects the locality of given by, Sm = R − m where R is the rank (in modulo
the correlations. 2 arithmetic) of gm [31]. The value of R can range from
There are several interesting directions that can be ex- m to 2m. When R = m, Sm = 0 implying that this
plored going forward. In the space of stabilizer states and m-qubit set is disentangled from the rest of the system.
random quantum circuits, several novel phase transitions Conversely, if R > m, this set is entangled with at least
and critical states have been found. It would be inter- some part of the rest of the system.
esting to resolve structures of critical states and compare This procedure can be used to calculate total corre-
different universality classes. We mostly focused on 1D lations of any region A consisting of disjoint groups of
but it is straightforward to extend to higher dimensional qubits, where each group is labeled by an integer i. The
states where the spatial structures of correlations can be total correlations P(multipartite mutual information) are
richer, especially close to criticality [28–31]. Another av- defined as, IA = i Si − SA . Here SA is the entangle-
enue which we haven’t explored is the evolution of the ment entropy of the entire region A with respect to the
entanglement structure diagram with time. This would rest of the system while Si is the entanglement entropy
give us important insights and visualization of how quan- of group i with the entire rest of the system.
tum information gets scrambled under chaotic dynamics To calculate our entanglement diagram we begin by
from simple initial states. Although our method is best placing all of our qubits in a list called clusters. Note
suited for stabilizer states, it would be fruitful to come that in later iterations of the algorithm, each element in
up with an efficient method to extend the idea of entan- clusters can contain multiple qubits. We then follow
glement structure diagrams to non-stablilizer states. the steps below:

Step 1: Set n=1. This integer will keep track of the

VI. ACKNOWLEDGEMENTS cluster weight.

Step 2: Calculate entanglement entropy, S of each ele-

This material is based upon work supported by the ment in clusters. If S = 0 for an element, that element
National Science Foundation under Grant No. PHY- is disentangled from the rest of the system. We remove
2409403. We would like to acknowledge fruitful discus- such elements from clusters and include them in the fi-
sions with Chaoming Jian during this research’s forma- nal diagram as isolated elements. The list clusters now
tive stages. has Lr elements remaining. If Lr = 0, the process ends.
If Lr ̸= 0, we proceed to the next step.

Appendix A: Algorithmic details for producing the Step 3: Increase n by 1.

entanglement structure diagrams
Step 4: Form all possible combinations/sets of n ele-
ments from clusters. There are Lnr of these. Calculate


Here we extend the discussion from Sec. II, giving a
detailed description of how we can construct the entan- the total correlations, I of each of these n-element sets.
glement structure diagrams for a stabilizer state, starting If I = 0 for all sets, we go back to Step 3 and repeat. If
from an arbitrary list of stabilizer generators. As de- I ̸= 0 for at least one set, we proceed to Step 5. We refer
scribed in Sec. II, the key task is to identify w-clusters in to the I ̸= 0 sets as indivisble.
each iteration until all qubits have been assigned to de- Step 5: Remove the elements of the indivisible sets
coupled large clusters. Each w-cluster is constructed by from clusters. Combine the indivisible sets into
identifying sets where subsets of w elements share non- their disjoint unions, then add these unions back into
zero total correlations. clusters as new elements. These are the n-clusters cor-
In the stabilizer formalism, a pure quantum state of L responding to the current iteration of the entanglement
qubits is described by an L × 2L binary matrix. The L structure diagram. Return to Step 1.
linearly independent rows encode the Pauli strings which
form the stabilizer generators. Each generator can be
mi ni
written in the form, g = Πi=L i=1 Xi Zi . The Xi , Zi are
The process ends when the list clusters is empty and
Pauli operators for the ith qubit and mi , ni can be 0 or at that point, we have the entire entanglement structure
1. The quantum state is a simultaneous eigenstate of diagram.
these L linearly-independent stabilizer generators. This
representation is not unique, as the state is unchanged
by adding any row to another (mod 2). Appendix B: Scaling of computation time with
In order to form our clusters, we must calculate the en- system size L
tanglement entropy of groups of m qubits with the rest of
the system. We find this entropy by truncating the stabi- The properties of the many-body state determines the
lizer matrix, keeping only the columns corresponding to computational time for the entanglement structure dia-
these m qubits. The resulting reduced stabilizer matrix gram. In particular, these diagrams are much more costly
(gm ) is an L × 2m matrix. The entanglement entropy is to calculate for states which contain clusters with large
 8

w. The reason for this is two-fold: First, given L ele- be calculated in a time which is polynomial in the system
ments, finding
indivisible subsets of w elements requires size. When n ∝ L, however, the time scales as eL ln L .
L
calculating w entropies. Second, each of these entropy The algorithm to find the entanglement structure dia-
calculations requires computing the rank of a L × 2w ma- gram itself reveals the complexity and information scram-
trix, a task whose complexity scales with w. bling of a quantum state. Although volume law states
In Sec. IV we encountered two types of volume law have diverging entanglement entropy, the level of infor-
states: Those for which the minimal stabilizer weight, n, mation scrambling can be markedly different depending
is independent of system size, and those for which n ∝ L. on whether it is generated by unitary operators or by
For the former, the entanglement structure diagram can projective measurements.

[1] R. Horodecki, P. Horodecki, M. Horodecki, and [17] S. Watanabe, Information theoretical analysis of multi-
K. Horodecki, Quantum entanglement, Rev. Mod. Phys. variate correlation, IBM Journal of Research and Devel-
81, 865 (2009). opment 4, 66 (1960).
[2] M. Ma, Y. Li, and J. Shang, Multipartite entan- [18] A. Kumar, Multiparty quantum mutual information: An
glement measures: a review, Fundamental Research alternative definition, Phys. Rev. A 96, 012332 (2017).
[Link] (2024). [19] D. Bouwmeester, J.-W. Pan, M. Daniell, H. Wein-
[3] J. Eisert, M. Cramer, and M. B. Plenio, Colloquium: furter, and A. Zeilinger, Observation of three-photon
Area laws for the entanglement entropy, Rev. Mod. Phys. greenberger-horne-zeilinger entanglement, Phys. Rev.
82, 277 (2010). Lett. 82, 1345 (1999).
[4] A. S. Sørensen and K. Mølmer, Entanglement and ex- [20] H. J. Briegel and R. Raussendorf, Persistent entangle-
treme spin squeezing, Phys. Rev. Lett. 86, 4431 (2001). ment in arrays of interacting particles, Phys. Rev. Lett.
[5] O. Gühne, G. Tóth, and H. J. Briegel, Multipartite en- 86, 910 (2001).
tanglement in spin chains, New Journal of Physics 7, 229 [21] M. P. Fisher, V. Khemani, A. Nahum, and S. Vijay,
(2005). Random quantum circuits, Annual Review of Condensed
[6] D. A. Meyer and N. R. Wallach, Global entanglement in Matter Physics 14, 335 (2023).
multiparticle systems, Journal of Mathematical Physics [22] S. Choi, Y. Bao, X.-L. Qi, and E. Altman, Quantum er-
43, 4273 (2002), [Link] ror correction in scrambling dynamics and measurement-
pdf/43/9/4273/19183190/4273 1 [Link]. induced phase transition, Phys. Rev. Lett. 125, 030505
[7] P. Hyllus, W. Laskowski, R. Krischek, C. Schwem- (2020).
mer, W. Wieczorek, H. Weinfurter, L. Pezzé, and [23] Y. Li, X. Chen, and M. P. A. Fisher, Measurement-
A. Smerzi, Fisher information and multiparticle entan- driven entanglement transition in hybrid quantum cir-
glement, Phys. Rev. A 85, 022321 (2012). cuits, Phys. Rev. B 100, 134306 (2019).
[8] J. Eisert and H. J. Briegel, Schmidt measure as a tool [24] B. Skinner, J. Ruhman, and A. Nahum, Measurement-
for quantifying multiparticle entanglement, Phys. Rev. induced phase transitions in the dynamics of entangle-
A 64, 022306 (2001). ment, Phys. Rev. X 9, 031009 (2019).
[9] A. Sen(De) and U. Sen, Channel capacities versus entan- [25] A. Lavasani, Y. Alavirad, and M. Barkeshli,
glement measures in multiparty quantum states, Phys. Measurement-induced topological entanglement transi-
Rev. A 81, 012308 (2010). tions in symmetric random quantum circuits, Nature
[10] A. Wong and N. Christensen, Potential multiparticle en- Physics 17, 342 (2021).
tanglement measure, Phys. Rev. A 63, 044301 (2001). [26] C.-M. Jian, Y.-Z. You, R. Vasseur, and A. W. W. Ludwig,
[11] C. Artiaco, T. K. Kvorning, D. A. Chávez, L. Herviou, Measurement-induced criticality in random quantum cir-
and J. H. Bardarson, Universal characterization of quan- cuits, Phys. Rev. B 101, 104302 (2020), arXiv:1908.08051
tum many-body states through local information (2024), [[Link]-mech].
arXiv:2410.10971 [quant-ph]. [27] S. Sang and T. H. Hsieh, Measurement-protected quan-
[12] T. K. Kvorning, L. Herviou, and J. H. Bardarson, Time- tum phases, Phys. Rev. Research 3, 023200 (2021).
evolution of local information: thermalization dynamics [28] X. Turkeshi, R. Fazio, and M. Dalmonte, Measurement-
of local observables, SciPost Phys. 13, 080 (2022). induced criticality in (2+1)-dimensional hybrid quantum
[13] C. Artiaco, C. Fleckenstein, D. Aceituno Chávez, T. K. circuits, Phys. Rev. B 102, 014315 (2020).
Kvorning, and J. H. Bardarson, Efficient large-scale [29] P. Sierant, M. Schirò, M. Lewenstein, and X. Turkeshi,
many-body quantum dynamics via local-information Measurement-induced phase transitions in (d +1 ) -
time evolution, PRX Quantum 5, 020352 (2024). dimensional stabilizer circuits, Phys. Rev. B 106, 214316
[14] Y. Li, X. Chen, and M. P. A. Fisher, Quantum zeno effect (2022), arXiv:2210.11957 [[Link]-mech].
and the many-body entanglement transition, Phys. Rev. [30] A. Lavasani, Z.-X. Luo, and S. Vijay, Monitored quantum
B 98, 205136 (2018). dynamics and the kitaev spin liquid, Phys. Rev. B 108,
[15] M. Ippoliti, M. J. Gullans, S. Gopalakrishnan, D. A. 115135 (2023).
Huse, and V. Khemani, Entanglement phase transitions [31] V. Sharma, C.-M. Jian, and E. J. Mueller, Subsystem
in measurement-only dynamics, Phys. Rev. X 11, 011030 symmetry, spin-glass order, and criticality from random
(2021). measurements in a two-dimensional bacon-shor circuit,
[16] D. Gottesman, Stabilizer codes and quantum error cor- Phys. Rev. B 108, 024205 (2023).
rection (1997), arXiv:quant-ph/9705052 [quant-ph]. [32] E. Bianchi, L. Hackl, M. Kieburg, M. Rigol, and L. Vid-
 9

mar, Volume-law entanglement entropy of typical pure

quantum states, PRX Quantum 3, 030201 (2022).