National Science Review

RESEARCH ARTICLE XX: output, Year

doi: https://doi.org/10.1093/nsr/XXXX

INFORMATION SCIENCE

Networked dynamic systems with higher-order

interactions: stability versus complexity

Downloaded from https://academic.oup.com/nsr/advance-article/doi/10.1093/nsr/nwae103/7631350 by guest on 22 March 2024

Ye Wang1 , Aming Li1,2,∗ and Long Wang1,2,†

ABSTRACT
The stability of complex systems is profoundly affected by underlying structures, which are often

PT
modeled as networks where nodes indicate system components and edges indicate pairwise
interactions between nodes. However, such networks cannot encode the overall complexity of
networked systems with higher-order interactions among more than two nodes. Set structures

RI
provide a natural description of pairwise and higher-order interactions where nodes are grouped into
multiple sets based on their shared traits. Here we derive the stability criteria for networked systems

SC
with higher-order interactions by employing set structures. In particular, we provide a simple rule
showing that the higher-order interactions play a double-sided role on community
stability—networked systems with set structures are stabilized if the expected number of common

U
sets for any two nodes is less than one. Moreover, although previous knowledge suggests that more
interactions (i.e., complexity) destabilize networked systems, we report that, with higher-order

N
interactions, networked systems can be stabilized by forming more local sets. Our findings are robust
with respect to degree heterogeneous structures, diverse equilibrium states and interaction types.
A
Keywords: networked systems, set structures, higher-order interactions, stability criteria
M
INTRODUCTION complex systems under different structures, lit-
ED

Explaining the intricate effect of community
structures on the stability of complex systems
has been a long-lasting challenge in mathemat-
IT
tle attention has been paid to higher-order in-
teractions with more than two species, namely,
the interaction between two species might be af-
fected by other species [30,31]. The hypergraph

ics, control, and ecology [1–11]. Stability, one

of the most important indicators in complex sys- formalism is a comprehensive framework for
studying higher-order interactions, which is fun-
ED

tems, portrays the ability of a system to return

to its equilibrium after external perturbations damentally equivalent to the set structure [32].
1
Center for Systems [12–17]. By analyzing the eigenvalue of the lin- Indeed, the set structure represents a typical
and Control, College of community structure that incorporates both pair-
earized community matrix, May first proves that
N

Engineering, Peking
University, Beijing for a random community where species interact wise and higher-order interactions. According
100871, China; randomly, achieving stability becomes increas- to ‘evolutionary set theory’ [33–35], species in
U

2
Center for Multi-Agent the realistic system are typically distributed over
Research, Institute for ingly challenging as community diversity and
Artificial Intelligence, complexity rise [1,2]. Following May’s frame- sets, where species within the same set interact
L

Peking University, work, both theoretical analysis and experiments with each other, and each species may belong
Beijing 100871, China to multiple sets based on multiple criteria. For
A

have shown that the underlying structure can sig-

nificantly affect the stability of complex systems, example, goldfish and lotus plants in the same
* Corresponding authors. pond could be considered members of the same
IN

Email:
such as the heterogeneous, multi-layer, temporal,
∗ amingli@pku.edu.cn, and higher-order structures, etc. [18–22]. set based on their shared location, but they would
† longwang@pku.edu.cn.
Previous studies generally use complex net- be classified into distinct sets based on their tax-
G

works to model various community structures, onomic differences. By grouping species into
Received: XX XX Year;
where nodes represent species and edges repre- multiple sets, set structures offer a more com-
RI

Revised: XX XX Year;
sent pairwise interactions [23–29]. Despite the prehensive perspective on species relationships
Accepted: XX XX Year
many deep insights gained about the stability of within a networked system. Indeed, previous
O

studies have demonstrated that the set structure

© The Author(s) 2024. Published by Oxford University Press on behalf of China Science Publishing & Media Ltd. This is an Open Access
article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which
permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.
 Natl Sci Rev, Year, Vol. XX, output
has a significant impact on the evolutionary dy-
namics of networked systems [33–35].
Nonetheless, compared with other exten-
sively studied network structures [30,36–38], the
effect of the set structure on the stability of
complex systems has been poorly understood.
Prior explorations have demonstrated that dif-
ferent network structures, such as heterogeneous
and modularity structures [39,40], could lead to
various stability criteria even under the same
interaction type. Based on the framework of
‘evolutionary set theory’ and stability theory
[12,33–35,41,42], here we investigate the impact
of the set structure on the stability of complex
systems under various interaction types, and an-
alyze the corresponding stability criteria in set
structured systems. To further understand the
role of set structures and higher-order interac-
tions, we also compare the stability of set struc-
tured systems with that of corresponding un-
structured (set-free) systems [4,18]. Our results
indicate that the number of sets each species be-
longs to (K) and the overall number of sets (G)
determine the stability of set structured systems.
Surprisingly, we find that the destabilizing effect
of increasing connectivity revealed by previous
studies [1,43] can be counteracted in set struc-
tured systems by reducing the expected num-
ber of common sets for two randomly selected
species (i.e., order, H = K 2 /G). Furthermore,
we provide a simple rule, H < 1, to enhance sta-
ED
bility in set structured systems, which is suitable
in various realistic scenarios such as heteroge-
neous structures, diverse equilibrium abundance,
IT
and diverse interaction types.
ED
RESULTS
Model
Here we employ the classical ordinary differen-
N
tial equations to describe the dynamics of species
interactions [4,18,43]
U
d𝑿 (𝑡)
= diag( 𝑿 (𝑡)) 𝒇 ( 𝑿 (𝑡)),
L
d𝑡
A
where 𝑿 (𝑡) = [𝑋1 (𝑡), 𝑋2 (𝑡), · · · , 𝑋𝑆 (𝑡)] T
contains the abundance of each species at time
IN
𝑡 and 𝒇 ( 𝑿 (𝑡)) captures the specific interac-
tions among species. The diagonal matrix
diag( 𝑿 (𝑡)) has the elements of 𝑿 (𝑡) on its di-
G
agonal and zeros elsewhere. We call that 𝑿 ∗ =
RI
[𝑋1∗ , 𝑋2∗ , · · · , 𝑋𝑆∗ ] T is a feasible equilibrium
point satisfying 𝒇 (𝑿 ∗ ) = 0 and 𝑋𝑖∗ > 0, 𝑖 =
1, 2, · · · , 𝑆.
O
Following the framework of stability theory,
we linearize the networked system to assess the
Page 2 of 11
stability of the system near 𝑿 ∗ , which subse-
quently gives
d𝒙(𝑡)
= diag 𝑿 ∗ 𝑱𝒙(𝑡) = 𝑴𝒙(𝑡),


(2)
d𝑡
𝜕𝒇
where 𝑱 = 𝜕𝑿 and 𝒙(𝑡) = 𝑿 (𝑡) − 𝑿 ∗ .
∗
diag ( 𝑿 ) 𝑱 is the community matrix of the sys-
tem, which is often denoted by 𝑴, whose ele-
ment 𝑀𝑖 𝑗 describes the effect that species 𝑗 has
Downloaded from https://academic.oup.com/nsr/advance-article/doi/10.1093/nsr/nwae103/7631350 by guest on 22 March 2024
on species 𝑖, and 𝑀𝑖𝑖 is the intrinsic effect of
species 𝑖. Mathematically, the stability of the
linearized system near the equilibrium is deter-
mined by the largest real part among all eigen-
values of 𝑴, Re(𝜆 𝑴 ,1 ). Specifically, a negative
PT
Re(𝜆 𝑴 ,1 ) indicates an asymptotically stable sys-
tem near the equilibrium, while the equilibrium
is unstable if Re(𝜆 𝑴 ,1 ) is positive.
RI
In our model, we consider the effect of set
structures on the stability of networked systems,
SC
where nodes represent species and nodes within
the same set can interact with each other. Let
G denote the number of sets, and each node be-
U
longs to K randomly chosen sets (K ≤ G), we
then construct the community matrix 𝑴 as fol-
N
lows: (i) for any species 𝑖 and set 𝑔, 𝑔𝑖 = 1 if
𝑖 belongs to 𝑔 and 𝑔𝑖 = 0 otherwise; (ii) for
A 𝑔 𝑔
any set 𝑔, we define (𝑀𝑖 𝑗 , 𝑀 𝑗𝑖 ) as the interac-
tion strength between species 𝑖 and 𝑗 in set 𝑔;
M
𝑔 𝑔
(iii) we draw (𝑀𝑖 𝑗 , 𝑀 𝑗𝑖 ) from a bivariate distri-
bution with probability 𝐶 if both species 𝑖 and
𝑗 are in set 𝑔 (𝑔𝑖 𝑔 𝑗 = 1), otherwise we set
𝑔 𝑔
𝑀𝑖 𝑗 = 𝑀 𝑗𝑖 = 0; (iv) for the community matrix
ÍG 𝑔
𝑴, 𝑀𝑖 𝑗 = 𝑔=1 𝑀𝑖 𝑗 ; (v) for diagonal elements
𝑀𝑖𝑖 , we set 𝑀𝑖𝑖 = −𝑑, where 𝑑 is a positive con-
stant.
In classical studies on the stability of com-
plex systems, the unstructured model supposes
that each pair of species interacts with proba-
bility 𝐶 (i.e., connectivity). The correspond-
ing community matrix 𝑴 ˆ of the unstructured
system primarily indicates pairwise interactions
[8,38] (Fig. 1a,d). These assumptions facili-
tate the theoretical analysis of complex systems
(1)
by focusing predominantly on pairwise interac-
tions within networked systems, which can be
denoted by edges. However, unstructured mod-
els cannot comprehensively capture the dynamic
nature of real-world systems, where species in-
teractions are not exclusively pairwise. In real-
ity, species are allowed to interact multiple times
since they may belong to various common sets
in the set structured
Í G system, whose community
matrix 𝑴 = 𝑔=1 𝑴 𝑔 incorporates both pair-
wise and higher-order interactions (Fig. 1b,c,e).
Indeed, the number of sets that any two species
have in common is defined as the order be-
 Natl Sci Rev, Year, Vol. XX, output

a b c
Unstructured community Species interactions in different sets Set structured community
with 6 species with 6 species

Set I
Set I Set II
√ √ √ √

Set II √ √ √ √

Set III √ √ √ √ Set III

Downloaded from https://academic.oup.com/nsr/advance-article/doi/10.1093/nsr/nwae103/7631350 by guest on 22 March 2024

d e
Unstructured community Set structured community
in large scale in large scale
Interaction Set
Species Number of sets: =4

PT
Number of sets each
species belongs to: =2
Number of species: S=34
Number of species: S=34

RI
Connectivity: C=0.1
Average connectivity
in each set: C=0.1

SC
Figure 1. Construction of the unstructured and set structured systems. (a) Interactions between six species in the unstructured
system. Each species interacts with others in the random network, where grey lines depict the interaction between each pair of

U
species. (b) The distribution of six species over three sets. Each species belongs to two sets randomly among the three sets,
and the expected number of species in each set is four. (c) Interactions among six species in the set structured system. Each

N
species interacts with others within the same set with probability C , where colored lines correspond to specific interactions in each
set. (d) A large-scale unstructured (random) system. Each species is represented by a blue circle, and grey edges represent
interactions between species. Here the number of species S = 34 and connectivity C = 0.1. (e) A large-scale set structured

A
system. Each species (blue circle) belongs to two sets among four sets. Species within the same set interact with probability C
(grey edges). Except for the setting of sets, other parameters in the set structured system are identical to those in the unstructured
M
system (number of species, connectivity, etc.).
ED

tween the two species. And the expected num-
ber of common sets for two randomly chosen
species (H ) is the overall order of the set struc-
tured system. Different from the classical net-
IT
two species compete for limited resources, their
interactions are competitive. Conversely, when
two species cooperate for mutual benefit, their
interactions are mutualistic.

worked systems [37,44], here we explore the sta-

bility of set structured systems and reveal the ef-
ED

fect of the combination of pairwise and higher- Stability criteria for diverse interaction types
order species interactions on community stabil- In contrast to classical unstructured (set-free)
ity. And for interactions between species, we systems where species interact with probabil-
N

mainly consider the following four types [4,18] ity 𝐶 [43,45], here we study the stability of set
(see Supplementary information S1 for details): structured systems and investigate the effect of
U

(i) Random. Interaction strengths between two sets on community stability. We compare the
species are drawn from a normal distribution largest real part among all eigenvalues of 𝑴 and
L

𝑁 (0, 𝜎 2 ) independently. (ii) Exploitative (+/−). 𝑴,ˆ which is the community matrix of the cor-
Interaction strengths between two species have responding unstructured (set-free) systems. To
A

opposite signs. One is drawn from the dis- ensure a fair comparison, the parameters of both
tribution −|𝑁 (0, 𝜎 2 )|, and the other is drawn set structured and unstructured systems are set to
IN

independently from the distribution |𝑁 (0, 𝜎 2 )|. be identical, such as the number of species (𝑆),
(iii) Mutualistic (+/+). Interaction strengths be- the connectivity (𝐶), and the interaction types.
G

tween two species have positive signs and are Theoretically, we provide the stability criteria to
drawn from |𝑁 (0, 𝜎 2 )| independently. (iv) Com- estimate the stability of set structured systems.
RI

petitive (−/−). Interaction strengths between Since a negative Re(𝜆 𝑴 ,1 ) indicates a stable sys-
two species have negative signs and are drawn tem, we use −Re(𝜆 𝑴 ,1 ) and −Re(𝜆 𝑴ˆ ,1 ) to rep-
from −|𝑁 (0, 𝜎 2 )| independently. These interac-
O

resent the stability of set structured and unstruc-

tion types can effectively describe complex real- tured systems, respectively [18,46].
world interaction scenarios. For example, when We find that, compared with the unstructured
Page 3 of 11
 Natl Sci Rev, Year, Vol. XX, output

a 0.76
b -0.3
e
1.0 +/-
Set structured -1.3
0.61 Random

Stability
-2.3 -/-
0.46 -3.3
Unstructured -4.3 -0.5
Random Mutualitic (+/+)
0.31 -5.3
c 0.92
d 0.62
-2.0
Stability 0.86 0.48

0.80 0.34 Simulation

Downloaded from https://academic.oup.com/nsr/advance-article/doi/10.1093/nsr/nwae103/7631350 by guest on 22 March 2024

Exploitative (+/-) Competitive (-/-) Theortical prediction +/+
0.74 0.20 -3.5
100 140 180 220 260 300 100 140 180 220 260 300 100 140 180 220 260 300
Number of species, S
Figure 2. Set structures improve community stability. (a-d) The stability of set structured (blue) and corresponding unstructured
(orange) systems with an increasing number of species, S . Specific interaction types in each panel correspond to random, mutual-

PT
istic (+/+) , exploitative (+/−) , and competitive (−/−) , respectively. (e) Theoretical predictions (lines) and numerical simulations
(dots) on the stability of set structured systems under different interaction types (different colors). We set the number of sets G = 50,
the number of sets each species belongs to K = 6, connectivity C = 0.5, σ = 0.05, and vary the number of species S from 100 to

RI
300. Each simulation result (box chart and dot) is obtained from over 50 replicates.

Table 1. Stability criteria for different interaction types in set structured and unstructured systems

SC
Stability criterion Stability criterion
Interaction type
(set structured community) (unstructured community)

U
√︃
2 √
Random 𝑆𝐶 KG < 𝑑 𝑑

N
𝜎 𝑆𝐶 < 𝜎

Mutualistic (𝑆 − 1) 𝐶 KG
2 √︁
2/𝜋 <
A
𝑑
𝜎
√︁
(𝑆 − 1) 𝐶 2/𝜋 < 𝑑
𝜎
M
√︃
2 √
Exploitative 𝑆𝐶 KG (1 − 2/𝜋) < 𝑑
𝜎 𝑆𝐶 (1 − 2/𝜋) < 𝑑
𝜎
ED

2 2 √︁
𝐹 (𝐶 KG ) + 𝐶 KG
√︁
Competitive 2/𝜋 < 𝑑
𝜎 𝐹 (𝐶) + 𝐶 2/𝜋 < 𝑑
𝜎
IT
ED

system, the set structure has a profound effect on General rule to stabilize complex
stabilizing the networked system if the number communities
of sets is much larger than the number of sets that Furthermore, we give a quantitative condition
N

each species belongs to (Fig. 2a-d), regardless of under which stability is enhanced in set struc-
the specific interaction types (random, exploita- tured systems, compared with that in unstruc-
U

tive, mutualistic, and competitive). tured systems. The expected number of com-
Next, we give the corresponding stability cri- mon sets, H , for two randomly chosen distinct
L

terion for different interaction types in set struc- species is (see Supplementary information S2.1
tured systems in Table 1 [4,47–49] (see methods
A

for details)
and Supplementary information S2 for details). K2
We verify our theoretical analysis of the stability H= . (3)
IN

criteria in set structured systems through Monte
Carlo simulations, and our theoretical predic-
G
G
Considering that species within the same set in-
teract at probability 𝐶, the expected interaction

tions are in good agreement with the numerical

results (Fig. 2e and Supplementary Fig. S1). times, 𝑇, for any two species is thus 𝑇 = 𝐶H =
2
RI

𝐶 KG . According to Table 1, we find that 𝑇 is a

variable that describes ‘complexity’ in set struc-
tured systems, corresponding to the connectivity
O

𝐶 in unstructured systems. Intuitively, increas-

ing connectivity 𝐶 decreases the stability of net-
Page 4 of 11
 Natl Sci Rev, Year, Vol. XX, output

a S=200 b S=400 c S=800

Stability (unstructured)
Stabilization Random

0 0 0
Destabilization
=1

Exploitative (+/-)

0 0 0

=1
Stability = Stability (set structured)

Downloaded from https://academic.oup.com/nsr/advance-article/doi/10.1093/nsr/nwae103/7631350 by guest on 22 March 2024

Mutualistic (+/+)
0 0 0

=1

PT
Competitive (-/-)
0 0 0

RI
0.36 0.53 1.00 3.60 0.36 0.53 1.00 3.60 0.36 0.53 1.00 3.60

SC
The expected number of sets any two species have in common,

Figure 3. Confirmation of the simple rule, K 2 /G < 1, to enhance stability in set structured systems. (a-c) The ∆Stability of

U
set structured and corresponding unstructured systems with increasing K 2 /G in small (S = 200 ) , middle (S = 400 ) , and large
size (S = 800 ) communities, respectively, where ∆Stability =Stability (set structured) − stability (unstructured). Interaction types

N
from top to bottom panels are random, exploitative (+/−) , mutualistic (+/+) , and competitive (−/−) , respectively. We highlight the
prediction of the critical ( K 2 /G) ∗ under different interaction types. The simple rule accurately predicts the critical ( K 2 /G) ∗ = 1
in random, exploitative, and mutualistic communities. The results suggest that K 2 /G < 1 is necessary but not sufficient in the

A
competitive community, as the critical ( K 2 /G) ∗ < 1 when the interaction type is competitive. For set structured systems, we set
the number of sets each species belongs to K = 6, connectivity C = 0.5, σ = 0.05, and vary the number of sets G from 10 to 100.
M
For unstructured systems, we employ the same parameters (C , σ ) as those in set structured cases. Each result is obtained from
over 50 replicates.
ED

worked systems since it increases the complex- stability in set structured systems, which is K 2 <
ity of the system [1,2]. Following May’s semi- G.
nal framework, species interact more frequently Our theoretical predictions are consistent
IT

with increasing 𝑇 in set structured systems, mak- with Monte Carlo simulations when the interac-
ing the system more complex. Our results show tion types are random, exploitative, or mutual-
ED

that the stability of set structured systems de- istic. We find that when the interaction type in
creases with increasing 𝑇, which increases the set structured systems is competitive, the critical
complexity (the expected interaction times for condition is more strict, and only a larger G can
N

any two species) of the whole system. stabilize the set structured system since the sta-
Surprisingly, we find that, the stability is neg- bility criterion is more sensitive to the varying G
U

atively correlated with the expected number of in competitive cases (Fig. 3a-c). Our findings in-
common sets H in set structured systems for dicate that higher-order species interactions (H )
random, exploitative, or mutualistic interactions.
L

have a double-sided role in community stability,

Compared with the corresponding unstructured and the stability is enhanced only if H < 1. And
A

systems with the same coefficients, the set struc- these results are also applicable in more general
tured systems are more stable (i.e., −Re(𝜆 𝑴ˆ ,1 )< cases, such as mixed interaction types (Supple-
IN

−Re(𝜆 𝑴 ,1 )) if H < 1, namely mentary Fig. S5) and various community densi-
ties (Supplementary Fig. S2).
K2
G

< 1. (4) To showcase the usefulness of our method

G and to provide quantitative comparisons with ex-
RI

That means, in set structured systems, species isting work [4], we next generate a significant
may form more local sets to overcome the desta- amount of simulated data to validate the sta-
bilizing effect of set structures on community
O

bilization effect of the increasing number of in-

teractions among species. Specifically, we pro- stability (see Supplementary information S3 for
vide a critical mathematical condition to enhance details). Our simulations track the dynamical
Page 5 of 11
 Natl Sci Rev, Year, Vol. XX, output

a Random b Exploitative (+/-) c Mutualistic (+/+) d Competitive (-/-)

2 Perturbation Stable Stable Stable Stable

Set structured
Recovery time Recovery time Recovery time Recovery time
1
Species

Abundance
2 Untable Stable Untable Untable

Unstructured
Recovery time
1

Downloaded from https://academic.oup.com/nsr/advance-article/doi/10.1093/nsr/nwae103/7631350 by guest on 22 March 2024

0 4 8 12 16 0 4 8 12 16 0 4 8 12 16 0 4 8 12 16
Time Time Time Time
e Small community (S=20) f Large community (S=500)
Set structured 98%

PT
100% 100% 100% 100% 98% 100% 100%
1.0
Unstructured
Proportion of stable

0.8
communities

64%

RI
0.6

0.4 32%
27%

SC
0.2
0% 0% 0%
0.0
Random +/- +/+ -/- Random +/- +/+ -/-

U
Interaction type

N
Figure 4. Simulated data verifies the stabilizing effect of set structures on community stability. (a-d) Simulation results
illustrate the evolution of species abundance over time in set structured and corresponding unstructured systems with the same
perturbations. We find that set structures exhibit a significant stabilizing effect on community stability if K 2 /G < 1, irrespective

A
of specific interaction types, such as random, exploitative (+/− ), competitive ( −/− ), and mutualistic (+/+) interactions (from left to
right panels). Even though both set structured and unstructured systems with exploitative (+/− ) interactions return to equilibrium
M
after perturbations, set structured systems show a notably shorter recovery time. Parameters: S = 10, σ = 0.5, C = 0.25, G =
6, K = 2. (e-f) The proportion of stable communities among 1000 simulated set structured and corresponding unstructured
communities. In small communities (S = 20), both set structured and unstructured systems tend to maintain stability. However, in
ED

large communities (S = 500), set structured systems sustain stability, while their unstructured counterparts grapple with instability.
Parameters: σ = 0.25, C = 0.25, G = 200, K = 2.
IT

change of species abundance in both set struc- the same set interact randomly. However, re-
tured and unstructured systems following identi- lated studies have shown that the topological
cal external perturbations. These numerical ex- structures in empirical systems can also be non-
ED

periments reveal that if the theoretical condition random and exhibit degree heterogeneity [50,
K 2 /G < 1 is satisfied, the relative abundance 51], indicating that species’ degree distributions
of each species converges more rapidly in set vary across different levels of generality. Specif-
N

structured systems than in unstructured systems, ically, degree heterogeneity refers to the un-
even when both systems are stable. And the set even distribution of connections of all nodes.
U

structured system can still be stable when the This concept signifies that within a network,
corresponding unstructured systems are unsta- some nodes possess a high degree of connectiv-
L

ble for K 2 /G < 1 (Fig. 4a-d). In addition, we ity, whereas others exhibit a low degree. For
also calculate the proportion of stable commu- example, in transportation networks like air-
A

nities among numerous simulated communities line routes, certain airports serve as major hubs
IN

and show that set structured systems are more with numerous direct flights to various destina-
likely to be stable in large complex communities tions, while smaller airports have fewer connec-
if G ≫ K 2 (Fig. 4e,f). tions. Next, we demonstrate that our frame-
G

work may analyze degree heterogeneous net-

works (see methods for details) and investigate
RI

Effect of degree heterogeneity on the effect of degree heterogeneity on community

community stability stability of set structured systems (Fig. 5e,f).
O

We have investigated the effect of set structures Remarkably, our conclusions still hold in de-
on community stability where species within gree heterogeneous networks, indicating that set

Page 6 of 11
 Natl Sci Rev, Year, Vol. XX, output

a b c d
Random Exploitative (+/-) Mutualistic (+/+) Competitive (-/-)

Stability
Stabilization
0 0
0
Destabilization 0

0.36 0.53 1.00 3.60 0.36 0.53 1.00 3.60 0.36 0.53 1.00 3.60 0.36 0.53 1.00 3.60
The expected number of sets any two species have in common,

Downloaded from https://academic.oup.com/nsr/advance-article/doi/10.1093/nsr/nwae103/7631350 by guest on 22 March 2024

e f
Scale-free community Set structured scale-free community

PT
RI
Figure 5. Confirmation of the simple rule, K 2 /G < 1, in heterogeneous systems. (a-d) The ∆Stability of set structured

SC
and corresponding unstructured systems with increasing K 2 /G in degree heterogeneous communities under random, exploitative
(+/−) , mutualistic (+/+) , and competitive (−/−) interaction types (from left to right panels). We highlight the critical ( K 2 /G) ∗ ≈
1 in degree heterogeneous communities. (e) Degree heterogeneous (scale-free) system. Each species is represented by a blue

U
circle, and grey edges represent interactions between species. Here the number of species S = 34 and average connectivity
C = 0.2. (f) Set structured degree heterogeneous (scale-free) system. Each species (blue circle) belongs to two sets among four

N
sets, and other corresponding parameters are identical to those in the unstructured system (number of species, average connectivity,
etc.). In (a-d), we set the connectivity C = 0.2, and other parameters are the same as those in Fig. 3a. For unstructured systems,
we employ the same parameters (C , σ ) as those in set structured cases. Each result is obtained from over 50 replicates.
A
M
structures play a dominant role in community dance [52,53], to verify the robustness of our
stability. Although degree heterogeneity can findings.
lead to fluctuating results, we show that the crit- Here we investigate the effect of local sym-
ED

∗
ical K 2 /G is consistently around 1 (Fig. 5a-
d), irrespective of specific interaction types (ran-
dom, exploitative, mutualistic, and competitive).
IT
metric structures with heterogeneous degree dis-
tributions on the stability of set structured sys-
tems. Specifically, the average degree in unstruc-

Moreover, our results are robust for mixed inter- tured systems is equivalent to that in each set in
action types (Supplementary Fig. S6). set structured systems (see methods for details).
ED

Following May’s framework, related studies And in local symmetric structures, the complex-
show that increasing connectivity (‘complexity’) ity of set structured systems is higher than that of
decreases stability in networked systems, regard- unstructured systems, as the overall average de-
N

less of the degree distribution of the system [4]. gree is higher than the average degree in each set
However, our results reveal that increasing the in set structured systems. Surprisingly, despite
U

number of sets G overcomes the destabilization the higher complexity of set structured systems
effect of increasing connectivity in set struc- in local symmetric structures, set structured sys-
tured systems in both degree homogeneity and tems can still be more stable than unstructured
L

heterogeneity networks. And the general rule systems if G ≫ K 2 . And under different inter-
A

∗
K 2 /G < K 2 /G ≈ 1 is a critical condition to


actions (random, exploitative, mutualistic,

∗ and
enhance stability in set structured systems com- competitive), the critical K 2 /G varies in lo-
IN

pared with unstructured systems. cal symmetric structures (Fig. 6a-d).

Moreover, we explore the effect of equilib-
rium abundance on the stability of set structured
G

Model extension and unstructured systems. In realistic systems,

RI

After exploring the effect of degree heterogene- species may have different equilibrium abun-
ity on the stability of set structured and un- dances, and previous studies have shown that
structured systems, we extend our models to these differences can significantly affect commu-
O

more complex scenarios, including local sym- nity stability. In previous sections, we postulate
metric structures and diverse equilibrium abun- that the equilibrium abundance 𝑋𝑖 ∗ = 1, 𝑖 =
Page 7 of 11
 Natl Sci Rev, Year, Vol. XX, output

a Random b Exploitative (+/-) c Mutualistic (+/+) d Competitive (-/-)

Local symmetric
0
Stabilization 0
0

structure
Destabilization 0

Stability
0.01 0.02 0.36 0.01 0.02 0.36 0.01 0.02 0.36 0.01 0.02 0.36
e f g h

Diverse equilibrium

Downloaded from https://academic.oup.com/nsr/advance-article/doi/10.1093/nsr/nwae103/7631350 by guest on 22 March 2024

0 0 0
0

abundance
0.36 0.53 1.00 3.60 0.36 0.53 1.00 3.60 0.36 0.53 1.00 3.60 0.36 0.53 1.00 3.60

PT
The expected number of sets any two species have in common,
Figure 6. Forming more local sets can improve the stability in local symmetric structures and with diverse equilibrium

RI
abundances. The ∆Stability of set structured and corresponding unstructured systems with increasing K 2 /G in local symmetric
communities (a-d) and with diverse equilibrium abundance in homogeneous communities (e-h) under random, exploitative (+/−) ,
mutualistic (+/+) , and competitive (−/−) interaction types (from left to right panels). We highlight the critical ( K 2 /G) ∗ ≈ 1 among

SC
all panels. (a-d) The average degree in both set structured and unstructured systems k = 4. (e-h) The equilibrium abundance is
drawn uniformly from the interval [ 0.75, 1.25 ] . We set parameters identical to those in Fig. 3a, and vary the number of sets G from
100 to 1000 (a-d) or 10 to 100 (e-h). Each result is obtained from over 50 replicates.

U
N
1, 2, · · · , 𝑆. Here we extend this assumption within the same set, potentially encompassing
by allowing for diverse equilibrium abundances interactions of any order within the community.
and investigate their effect on the stability of
set structured and unstructured systems in both A
This general and innovative approach promises
to greatly enhance our comprehension of intri-
M
homogeneous and heterogeneous communities.
∗ cate community dynamics. By incorporating
Our findings reveal that the critical K 2 /G re- both pairwise and higher-order interactions, set
mains close to 1, regardless of the diverse equi- structures effectively capture the complexity of
ED

librium abundances (Fig. 6e-h). We also ex- interactions in networked systems.

amine different equilibrium abundance distribu- Indeed, as a kind of hypergraph [32], set
tions, such as log-normal and half-normal distri- structures are ubiquitous in both nature and
IT

butions, and find that they do not qualitatively al- human society, from microorganisms and ani-
ter our results (Supplementary Fig. S3). We fur- mals to human populations [33,34]. Specifi-
ED

ther consider degree heterogeneous communities

∗ cally, species may belong to multiple sets si-
and demonstrate that the critical K 2 /G is still multaneously under different classification crite-
near 1 under diverse equilibrium abundances in ria, playing different roles and functions in each
such communities (Supplementary Fig. S4, S7- set. For example, whales are both marine and
N

S9). mammalian in the sea system, while sharks are

marine but not mammalian. Previous studies
U

demonstrate that set structures qualitatively in-

DISCUSSION fluence the evolutionary outcomes of complex
L

Higher-order interactions are of paramount im- systems [34,35]. Here we consider complex sys-
A

portance in various fields such as network sci- tems with higher-order interactions by introduc-
ence, ecology, mathematics, and sociology [30– ing set structures, and investigate the stability of
IN

32]. In particular, extensive research has been set structured systems. We establish the stabil-
devoted to understanding the impact of higher- ity criteria in set structured systems. Our find-
order interactions on complex systems by con- ings demonstrate that, contrary to the classical
G

sidering third or fourth-order interactions [30, ‘diversity-stability debate’ [1,2], increasing the
31]. Our study introduces set structures, equiv- number of sets, G, has a stabilizing effect on
RI

alent to hypergraphs [32], as a novel framework set structured systems, which can overcome the
for analyzing higher-order species interactions. destabilization effect of increasing connectivity.
O

Within this framework, species are organized Surprisingly, we find that, compared with the
into distinct sets and interact with other species corresponding unstructured systems, there exists

Page 8 of 11
 Natl Sci Rev, Year, Vol. XX, output
a simple general rule to stabilize set structured
systems. Moreover, we have theoretically de-
rived that the set
structured system

∗ is more sta-
ble when K 2 /G < K 2 /G ≈ 1, irrespective
of interaction types. Furthermore, we find that
set structures play a significant role in commu-
nity stability, and the effect of set structures on
system stability is robust under various parame-
ters, including the number of species, connectiv-
ity, etc. Remarkably, we also discuss the effect of
degree heterogeneous structures, diverse equilib-
rium abundance, and mixed interaction types on
the stability of set structured systems. And our
rule still holds when considering these realistic
factors. Overall, our study provides a new per-
spective on understanding the role of set struc-
tures in community stability and offers a general
framework to study the role of set structures in
complex systems.
Our model employs set structures as a frame-
work to investigate the impact of higher-order
interactions on the stability of networked dy-
namic systems. It is worth noting that there
exist various ways to characterize higher-order
interactions, such as hyperedges and simplicial
complexes. Therefore, the exploration of multi-
ple definitions of higher-order interactions holds
the potential for a comprehensive examination
M
of their impact on the stability and complex-
ity of networked systems. Additionally, the sta-
bility of complex systems in the real world is
ED
subject to various factors such as stochasticity,
nonlinearity, and reactivity [46,54–57]. Future
work based on our framework can further con-
IT
sider these factors, contributing to a more thor-
ough understanding of complex systems. Fur-
thermore, exploring real-world networks with set
ED
structures and diverse species interactions repre-
sents a promising direction for future research
based on our theoretical findings, with the key
N
challenge being the acquisition of relevant data.
Therefore, efforts to obtain and analyze such
U
data can greatly enhance our understanding of
the stability of empirical systems. In addition,
L
previous work has pointed out that environmen-
tal feedback plays a pivotal role in shaping the
A
dynamics of complex systems [56]. Combining
environmental feedback with higher-order inter-
IN
actions can extend our work into the broader
context of co-evolutionary dynamics research,
G
deepening our understanding of the impact of
higher-order interactions on the stability of net-
RI
worked dynamic systems.
Another potential extension points to the fu-
O
ture application of our model in more realis-
tic scenarios, such as multi-layer and time-delay
Page 9 of 11
systems [18,46]. By investigating the role of
higher-order interactions in these realistic com-
plex systems, we can better understand the sys-
tem stability. This exploration also highlights
that higher-order interactions impact not only the
stability and complexity of complex systems but
also their fundamental functions. For example,
by identifying the sets most critical to commu-
nity stability, we can prioritize conservation ef-
Downloaded from https://academic.oup.com/nsr/advance-article/doi/10.1093/nsr/nwae103/7631350 by guest on 22 March 2024
forts and develop targeted interventions to pro-
tect those sets. These extensions will help us
understand more about the system stability and
have profound theoretical and practical implica-
tions, which also significantly enrich the investi-
gation of the effect of set structures on the stabil-
PT
ity of complex systems.
RI
METHODS
Heterogeneous community matrix
SC
In each set 𝑔, (𝑔 = 1, 2, · · · , G), we construct
a scale-free network 𝑮 𝒈 with connectivity 𝐶,
where nodes represent species and edges capture
U
interactions. The construction process of scale-
N
free networks typically follows the preferential
attachment mechanism [50], where new nodes
A
are more likely to connect to existing nodes with
high degrees. If species 𝑖 and 𝑗 are in the same
set 𝑔 (𝑔𝑖 = 𝑔 𝑗 = 1) and there exists an edge be-
𝑔 𝑔
tween them, we sample (𝑀𝑖 𝑗 , 𝑀 𝑗𝑖 ) from a given
𝑔 𝑔
bivariate distribution, otherwise 𝑀𝑖 𝑗 = 𝑀 𝑗𝑖 = 0.
Different interaction types correspond to distinct
bivariate distributions. And we have
G
∑︁
𝑔
𝑀𝑖 𝑗 = 𝑀𝑖 𝑗 . (5)
𝑔=1
In the corresponding heterogeneous unstruc-
tured systems, we first construct a scale-free net-
work 𝑮 with connectivity 𝐶. And for any species
𝑖 and 𝑗, we sample ( 𝑀ˆ 𝑖 𝑗 , 𝑀ˆ 𝑗𝑖 ) from the same
distribution as that in heterogeneous set struc-
tured systems if 𝐺 𝑖 𝑗 = 𝐺 𝑗𝑖 = 1, otherwise
𝑀ˆ 𝑖 𝑗 = 𝑀ˆ 𝑗𝑖 = 0.
Local symmetric structure community
matrix
In each set 𝑔, we construct a scale-free network
𝑮 𝒈 whose average degree is 𝑘. If 𝑔𝑖 𝑔 𝑗 = 1 and
there exists an edge between species 𝑖 and 𝑗, we
𝑔 𝑔
sample (𝑀𝑖 𝑗 , 𝑀 𝑗𝑖 ) from a given bivariate distri-
𝑔 𝑔
bution, otherwise 𝑀𝑖 𝑗 = 𝑀 𝑗𝑖 = 0. Then, we have
ÍG 𝑔
𝑀𝑖 𝑗 = 𝑔=1 𝑀𝑖 𝑗 .
In the corresponding unstructured systems,
we also construct a scale-free network 𝑮 whose
 Natl Sci Rev, Year, Vol. XX, output

average degree is 𝑘. And we sample ( 𝑀ˆ 𝑖 𝑗 , 𝑀ˆ 𝑗𝑖 ) Data and code availability

from the same distribution as that in set struc-
tured systems if 𝐺 𝑖 𝑗 = 𝐺 𝑗𝑖 = 1, otherwise
𝑀ˆ 𝑖 𝑗 = 𝑀ˆ 𝑗𝑖 = 0.
Stability criteria
(1) Random. When interactions between species
are random,
for the community matrix
𝑴, we
All data analyzed can be reproduced us-
ing the codes provided, which are available
at https://github.com/yewang0109/NetStaCom.
The codes are written in MathWorks Matlab
R2022b.
Supplementary data

Downloaded from https://academic.oup.com/nsr/advance-article/doi/10.1093/nsr/nwae103/7631350 by guest on 22 March 2024

have E 𝑀𝑖 𝑗 𝑖≠ 𝑗 = 0 and Var 𝑀𝑖 𝑗 𝑖≠ 𝑗 = Supplementary data are available at NSR online.
K2 2
𝐶 G 𝜎 , where G is the number of sets and K is
the number of sets each species belongs to. The
Acknowledgement
above relations give the corresponding stability
criterion for random interactions as shown in Ta- We thank four anonymous referees for construc-

PT
ble 1 (see Supplementary information S2.2 for tive suggestions.
details)
(2) Exploitative (+/−). For the exploita-
Funding

RI
tive case that interactions between species
in
each set are exploitative, we have E 𝑀𝑖 𝑗 𝑖≠ 𝑗 = This work is supported by the National
Key Research and Development Program of

SC
2
= −𝐶 KG E2 (|𝑋 |) and


0, E 𝑀𝑖 𝑗 𝑀 𝑗𝑖 𝑖≠ 𝑗
2
China (2022YFA1008400), National Natu-
Var 𝑀𝑖 𝑗 𝑖≠ 𝑗 = 𝐶 KG Var (𝑋), where 𝑋 ∼


ral Science Foundation of China (62036002
𝑁 (0, 𝜎 2 ). The above relations give the corre- and 62173004), and Beijing Nova Program

U
sponding stability criterion for exploitative inter- (Z211100002121105).

N
actions as shown in Table 1 (see Supplementary
information S2.3 for details).
Author contributions
(3) Mutualistic (+/+). In the mutualistic case,
𝑔 𝑔
(𝑀𝑖 𝑗 , 𝑀 𝑗𝑖 ) are drawn from the distribution |𝑋 | A
All authors developed the model, performed the
M
with probability 𝐶 if species 𝑖 and 𝑗 both be- research, and wrote the paper.
long to the set ℎ. And for the community ma-
2 Conflict of interest statement. None declared.
trix 𝑴, we have E 𝑀𝑖 𝑗 𝑖≠ 𝑗 = 𝐶 KG E (|𝑋 |) and


ED

2
 2
2
Var 𝑀𝑖 𝑗 𝑖≠ 𝑗 = 𝐶 KG E 𝑋 2 − 𝐶 KG E (|𝑋 |) ,



REFERENCES
where 𝑋 ∼ 𝑁 (0, 𝜎 2 ). The above relations give
IT

1. May RM. Will a large complex system be stable? Nature

the corresponding stability criterion for mutual- 1972; 238: 413–4.
istic interactions as shown in Table 1 (see Sup- 2. May RM. Stability and complexity in model ecosystems.
ED

plementary information S2.4 for details). Princeton, NJ: Princeton University Press, 2001.
(4) Competitive (−/−). We finally dis- 3. Šiljak DD. Large-scale dynamic systems: stability and
cuss the competitive case where species structure, New York: North-Holland, 1978.
N

interact competitively. Similarly, we 4. Allesina S and Tang S. Stability criteria for complex ecosys-
2
−𝐶 KG E (|𝑋 |) and

tems. Nature 2012; 483: 205-8.
have E 𝑀𝑖 𝑗 𝑖≠ 𝑗 =
U

2 5. Araki M. Stability of large-scale nonlinear systems–

E 𝑀𝑖 𝑗 𝑀 𝑗𝑖 𝑖≠ 𝑗 = 𝐶 KG E2 (|𝑋 |). The above


quadratic-order theory of composite-system method using
relations give the corresponding stability cri- M-matrices. IEEE Trans Automat Contr 1978; 23: 129–42.
L

terion for competitive interactions as shown in 6. Donkers MCF, Heemels WPMH, van de Wouw N et al.
Table 1 (see Supplementary information S2.5 Stability analysis of networked control systems using a
A

for details), where 𝐹 (𝑥) satisfies that switched linear systems approach. IEEE Trans Automat
Contr 2011; 56: 2101–15.
IN

√︂   
𝑥 2 (1 − 𝑥) 7. Donkers MCF, Heemels WPMH, Bernardini D et al. Stability
𝐹 (𝑥) = 𝑆𝑥 1 − 2 1+ . (6) analysis of stochastic networked control systems. Automat-
𝜋 𝜋 − 2𝑥
ica 2012; 48: 917–25.
G

Note that our stability criteria for competitive in- 8. Montoya JM, Pimm SL and Solé RV. Ecological networks
RI

and their fragility. Nature 2006; 442: 259–64.

teractions in set structured systems are obtained
9. Girvan M and Newman MEJ. Community structure in social
through the approximate estimation, which is vi-
2 and biological networks. Proc Natl Acad Sci USA 2002; 99:
olated under the condition that 𝐶 KG ≪ 1.
O

7821–6.
10. Estrada R. On the stability of multiloop feedback systems.
IEEE Trans Automat Contr 1972; 17: 781–91.
Page 10 of 11
 Natl Sci Rev, Year, Vol. XX, output
11. Michel AN. Stability analysis of interconnected systems.
SIAM J Control 1974; 12: 554–79.
12. Stuart JT. Nonlinear stability theory. Annu Rev Fluid Mech
1971; 3: 347–70.
13. Grujić LT and Šiljak DD. Asymptotic stability and instability
of large-scale systems. IEEE Trans Automat Contr 1973;
18: 636–45.
14. Grujić LT and Šiljak DD. On stability of discrete composite
systems. IEEE Trans Automat Contr 1973; 18: 522–4.
15. Kharitonov VL and Niculescu SI. On the stability of linear
systems with uncertain delay. IEEE Trans Automat Contr
2003; 48: 127–32.
16. Lasley EL and Michel AN. Input-output stability of inter-
connected systems. IEEE Trans Automat Contr 1976; 21:
84–9.
17. Michel AN and Rasmussen RD. Stability of stochastic
composite systems. IEEE Trans Automat Contr 1976; 21:
89–94.
18. Wang Y, Yang Y, Li A et al. Stability of multi-layer ecosys-
tems. J R Soc Interface 2023; 20: 20220752.
19. Pilosof S, Porter MA, Pascual M et al. The multilayer nature
of ecological networks. Nat Ecol Evol 2017; 1: 0101.
20. Levine JM, Bascompte J, Adler PB et al. Beyond pairwise
mechanisms of species coexistence in complex communi-
ties. Nature 2017; 546: 56–64.
21. Meena C, Hens C, Acharyya S et al. Emergent stability in
complex network dynamics. Nat Phys 2023; 19: 1033–42.
22. Barzel B and Barabási AL. Network link prediction by global
silencing of indirect correlations. Nat Biotechnol 2013; 31:
720–5.
23. Wang L and Xiao F. Finite-time consensus problems for
networks of dynamic agents. IEEE Trans Automat Contr
2010; 55: 950–5.
24. Montestruque LA and Antsaklis PJ. On the model-based
ED
control of networked systems. Automatica 2003; 39:
1837–43.
25. Nešić D and Teel A. Input-to-state stability of networked
control systems. Automatica 2004; 40: 2121–8.
IT
26. Newman MEJ. Modularity and community structure in net-
works. Proc Natl Acad Sci USA 2006; 103: 8577–82.
ED
27. Li A, Cornelius SP, Liu YY et al. The fundamental advan-
tages of temporal networks. Science 2017; 358: 1042–6.
28. Li A, Zhou L, Su Q et al. Evolution of cooperation on tem-
N
poral networks. Nat Commun 2020; 11: 2259.
29. Xiao F and Wang L. Asynchronous consensus in
continuous-time multi-agent systems with switching topol-
U
ogy and time-varying delays. IEEE Trans Automat Contr
2008; 53: 1804–16.
L
30. Bairey E, Kelsic ED and Kishony R. High-order species in-
teractions shape ecosystem diversity. Nat Commun 2016;
A
7: 12285.
31. Gibbs T, Levin SA and Levine JM. Coexistence in diverse
IN
communities with higher-order interactions. Proc Natl Acad
Sci USA 2022; 119: e2205063119.
32. Alvarez-Rodriguez U, Battiston F, de Arruda GF et al. Evo-
G
lutionary dynamics of higher-order interactions in social
networks. Nat Hum Behav 2021; 5: 586–95.
RI
33. Tarnita CE, Antal T, Ohtsuki H et al. Evolutionary dynam-
ics in set structured populations. Proc Natl Acad Sci USA
O
2009; 106: 8601–4.
34. Tarnita CE, Ohtsuki H, Antal T et al. Strategy selection in
structured populations. J Theor Biol 2009; 259: 570–81.
Page 11 of 11
35. Nowak MA, Tarnita CE and Antal T. Evolutionary dynamics
in structured populations. Phil Trans R Soc B 2010; 365:
19–30.
36. Nowzari C, Garcia E and Cortés J. Event-triggered commu-
nication and control of networked systems for multi-agent
consensus. Automatica 2019; 105: 1–27.
37. Tabbara M and Nešić D. Input–output stability of networked
control systems with stochastic protocols and channels.
IEEE Trans Automat Contr 2008; 53: 1160–75.
38. Baron JW and Galla T. Dispersal-induced instability in com-
Downloaded from https://academic.oup.com/nsr/advance-article/doi/10.1093/nsr/nwae103/7631350 by guest on 22 March 2024
plex ecosystems. Nat Commun 2020; 11: 6032.
39. Grilli J, Rogers T and Allesina S. Modularity and stability in
ecological communities. Nat Commun 2016; 7: 12031.
40. Fortuna MA, Stouffer DB, Olesen JM et al. Nestedness
versus modularity in ecological networks: two sides of the
same coin? J Anim Ecol 2010; 79: 811–7.
PT
41. Socha L. The asymptotic stochastic stability in large of
the composite stochastic systems. Automatica 1986; 22:
605–10.
RI
42. Silverman LM and Anderson BDO. Controllability, observ-
ability and stability of linear systems. SIAM J Control 1968;
6: 121–30.
SC
43. Allesina S and Tang S. The stability–complexity relationship
at age 40: a random matrix perspective. Popul Ecol 2015;
57: 63–75.
U
44. Strogatz SH. Exploring complex networks. Nature 2001;
410: 268–76.
N
45. Roberts A. The stability of a feasible random ecosystem.
Nature 1974; 251: 607–8.
A
46. Yang Y, Foster KR, Coyte KZ et al. Time delays modulate
the stability of complex ecosystems. Nat Ecol Evol 2023; 7:
M
1610–9.
47. Tao T, Vu V and Krishnapur M. Random matrices: univer-
sality of ESDs and the circular law. Ann Probab 2010; 38:
2023–65.
48. Sommers HJ, Crisanti A, Sompolinsky H et al. Spectrum
of large random asymmetric matrices. Phys Rev Lett 1988;
60: 1895–8.
49. Girko VL. Circular law. Theory Probab Appl 1985; 29:
694–706.
50. Barabási AL and Albert R. Emergence of scaling in random
networks. Science 1999; 286: 509–12.
51. Barzel B and Barabási AL. Universality in network dynam-
ics. Nat Phys 2013; 9: 673–81.
52. Ehrlén J and Morris WF. Predicting changes in the dis-
tribution and abundance of species under environmental
change. Ecol Lett 2015; 18: 303–14.
53. Gibbs T, Grilli J, Rogers T et al. Effect of population abun-
dances on the stability of large random ecosystems. Phys
Rev E 2018; 98: 022410.
54. Lenormand T, Roze D and Rousset F. Stochasticity in evo-
lution. Trends Ecol Evol 2009; 24: 157–65.
55. Sæther BE. Environmental stochasticity and population dy-
namics of large herbivores: a search for mechanisms.
Trends Ecol Evol 1997; 12: 143–9.
56. Wang X and Fu F. Eco-evolutionary dynamics with envi-
ronmental feedback: cooperation in a changing world. EPL
2020; 132: 1001.
57. Yang Y, Coyte KZ, Foster KR et al. Reactivity of com-
plex communities can be more important than stability. Nat
Commun 2023; 14: 7204.