See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/376028701

Hegselmann-Krause and Cucker-Smale type models with attractive-repulsive

interaction

Preprint · November 2023

DOI: 10.48550/arXiv.2311.13238

CITATIONS READS

0 72

2 authors:

Elisa Continelli Cristina Pignotti

Università degli Studi dell'Aquila Università degli Studi dell'Aquila
9 PUBLICATIONS 5 CITATIONS 79 PUBLICATIONS 2,333 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Elisa Continelli on 29 November 2023.

The user has requested enhancement of the downloaded file.

 Hegselmann-Krause and Cucker-Smale type models with

attractive-repulsive interaction

Elisa Continelli & Cristina Pignotti

Dipartimento di Ingegneria e Scienze dell'Informazione e Matematica
Università degli Studi di L'Aquila
Via Vetoio, Loc. Coppito, 67100 L'Aquila Italy
November 22, 2023

Abstract
In this paper, we analyze a Hegselmann-Krause opinion formation model and a Cucker-
Smale ocking model with attractive-repulsive interaction. To be precise, we investigate the
situation in which the individuals involved in an opinion formation or a ocking process
attract each other in certain time intervals and repeal each other in other ones. Under quite
general assumptions, we prove the convergence to consensus for the Hegselmann-Krause
model and the exhibition of asymptotic ocking for the Cucker-Smale model in presence
of positive-negative interaction. With some additional conditions, we are able to improve
the convergence to consensus for the solutions of the Hegselmann-Krause model, namely we
establish an exponential convergence to consensus result.

1 Introduction
Multiagent systems have been widely studied in these last years, due to their application to
several scientic elds, such as biology [8, 16], economics [26, 21], robotics [7, 22], control theory
[3, 6, 33, 28, 29, 2], social sciences [4, 31, 9, 1, 23]. Among them, there are the Hegselmann-
Krause opinion formation model [20] and its second order version, the Cucker-Smale ocking
model, introduced in [16] to describe ocking phenomena. For the solutions of such systems,
the asymptotic behaviour is usually investigated, namely, the convergence to consensus or the
exhibition of asymptotic ocking are analyzed for the Hegselmann-Krause model and the Cucker-
Smale model, respectively.
For the two aforementioned multiagent systems, dierent scenarios have been considered.
Among others, there is the situation in which the particles of the system interrupt their inter-
action at certain times. In [5], the convergence to consensus and the exhibition of asymptotic
ocking for a class of Cucker-Smale systems with communication failures have been proved under
suitable assumptions. With a dierent approach, the exponential convergence to consensus for
the solutions of Hegselmann-Krause type models involving a weight function that can degenerate
at some times has been achieved in [15], also in presence of time delays.
The Hegselmann-Krause model and the Cucker-Smale model in presence of time delays, but
in the not degenerate case, have been also analyzed by several authors (see e.g. [13, 14, 25, 24,
10, 11, 30, 19, 17, 12, 18, 32])

1
 A situation that to our knowledge has been not examined yet is the one in which the agents
have positive-negative interaction, namely the system's particles attract each other in certain
time intervals and repeal each other in other ones.
The aim of this paper is indeed to nd conditions for which the convergence to consensus and
the exhibition of asymptotic ocking are guaranteed for the solutions of the Hegselmann-Krause
model and of the Cucker-Smale model, respectively, with attractive-repulsive interaction. In this
case, in order to get the asymptotic consensus or the asymptotic ocking, one has to compensate
the behaviour of the solutions to the considered model in the bad intervals, i.e. the intervals in
which the agents repeal each other, with the good behaviour of the solutions of the system in the
intervals in which the inuence among the agents is positive. Under suitable assumptions, we are
able to prove the convergence to consensus for the Hegselmann-Krause model with attractive-
repulsive interaction and the exhibition of asymptotic ocking for the Cucker-Smale model with
attractive-repulsive interaction. Of course, this will require some restrictions on the length of the
bad time intervals. Under some additional conditions, it can be proved that the solutions of the
Hegselmann-Krause model in presence of positive-negative interaction converge exponentially to
consensus.
The paper is organized as follows. In Section 2, we deal with a Hegselmann-Krause model
with attractive-repulsive interaction in presence of a very general type of inuence function.
We give the denition of diameter for the solutions to the model and the rigorous denition of
convergence to consensus. Also, we state our main result, that ensures the asymptotic consensus
for our Hegselmann-Krause type model under quite general assumptions.
In Section 3, we present some preliminary results that are needed for the proof of our conver-
gence to consensus result while, in Section 4, we prove our main result. Also, under less general
assumptions, we provide an exponential convergence to consensus result.
In Section 5 we formulate the convergence to consensus for a Hegselmann-Krause model with
attractive-repulsive interaction in the particular case in which the inuence function is no longer
a general function of the opinions of the agents but a function of the distance of the agents'
opinions than can eventually tend to zero as the distance of the opinions becomes too large.
In Section 6, we consider a Cucker-Smale model in presence of attractive-repulsive interaction.
In this case, the inuence function cannot be a general function of the agents' opinion but must
rather be a function of the distance of the opinions of the system's particles. However, the
monotonicity of the inuence function, that is assumed in general when dealing with Cucker-
Smale type models, is removed (see also [13]). We give the denitions of the diameter in position,
the diameter in velocity, and the exhibition of asymptotic ocking for the solutions of the Cucker-
Smale model. Also, we state our main result, that ensures that every solution to the Cucker-
Smale model with attractive-repulsive interaction reaches the asymptotic ocking under suitable
assumptions.
Finally, in Section 7 we prove our ocking theorem through the use of some preliminary
results.

2 The Hegselmann-Krause model with attractive-repulsive inter-

action and general inuence function
Let us denote with N0 := N ∪ {0}. Consider a nite set of N ∈ N particles, with N ≥ 2. Let
xi (t) ∈ IRd be the opinion of the i-th individual at time t. We shall denote with |·| and h·, ·i the
usual norm and scalar product on IRd , respectively. The interactions between the elements of

2
the system are described by the following Hegselmann-Krause type model
d X
xi (t) = α(t)aij (t)(xj (t) − xi (t)), t > 0, ∀i = 1, . . . , N. (2.1)
dt
j:j6=i

with initial conditions

xi (0) = x0i ∈ Rd , ∀i = 1, . . . , N. (2.2)
Here, the communication rates aij are of the form
1
aij (t) := ψ(xi (t), xj (t)), ∀t > 0, ∀i, j = 1, . . . , N, (2.3)
N −1
where ψ : IRd × IRd → IR is a given inuence function. We assume that the inuence function ψ
is positive, continuous, and bounded and we denote
K := kψk∞ . (2.4)
Moreover, the weight function α : [0, +∞) → {−1, 1} is dened as follows
(
1, t ∈ (t2n , t2n+1 ) , n ∈ N0 ,
α(0) = 1, α(t) = (2.5)
−1, t ∈ [t2n+1 , t2n+2 ] , n ∈ N0 ,
where {tn }n is a sequence of nonnegative numbers such that t0 = 0, tn → ∞ as n → ∞ and
ln 2
t2n+2 − t2n+1 < , ∀n ∈ N0 . (2.6)
K
Also, we set
Mn0 := max |xi (tn )|. (2.7)
i=1,...,N

We dene the diameter d(·) of the system as follows:

d(t) := max |xi (t) − xj (t)|, ∀t ≥ 0.
i,j=1,...,N

Denition 2.1. We say that a solution {xi }i=1,...,N to system (2.1) converges to consensus if
lim d(t) = 0.
t→+∞

We will prove the following convergence to consensus result.

Theorem 2.1. Let ψ : IRd × IRd → IR be a positive, bounded, continuous function. Assume that
the sequence {tn }n of denition (2.5) satises (2.6). Assume also that the following conditions
hold: ∞
!
X eK(t2p+2 −t2p+1 )
ln < +∞, (2.8)
p=0
2 − eK(t2p+2 −t2p+1 )
∞   
X
−K(t2p+1 −t2p ) ψ0 −K(t2p+1 −t2p )
ln max 1 − e ,1 − (1 − e ) = −∞, (2.9)
K
p=0
where
ψ0 := min ψ(y, z), (2.10)
|y|,|z|≤M 0
being P∞
M 0 = eK p=0 (t2p+2 −t2p+1 ) M00 . (2.11)
Then, every solution {xi }i=1,...,N to (2.1) with the initial conditions (2.2) converges to consensus.

3
Remark 2.2. Let us note that condition (2.8) implies that
+∞
X
(t2p+2 − t2p+1 ) < +∞, (2.12)
p=0

so that the quantity M 0 in (2.11) is nite and it makes sense to consider the minimum given by
(2.10). Indeed, being eK(t2p+2 −t2p+1 ) > 1, it turns out that

eK(t2p+2 −t2p+1 )
eK(t2p+2 −t2p+1 ) ≤ ,
2 − eK(t2p+2 −t2p+1 )
from which
+∞ +∞
!
X X eK(t2p+2 −t2p+1 )
K (t2p+2 − t2p+1 ) ≤ ln .
p=0 p=0
2 − eK(t2p+2 −t2p+1 )

Then, from (2.8), the condition (2.12) is satised. As a consequence, t2p+2 − t2p+1 → 0, as
p → ∞.

Remark 2.3. Assume that

 
1 K
t2n+1 − t2n > ln 1 + , ∀n ∈ N0 .
K ψ0
 
Note that ln 1 + K
ψ0 > ln 2, so that

t2p+2 − t2p+1 < t2q+1 − t2q , ∀p, q ∈ N0 .

Then, in this situation the condition (2.9) can be simplied since

 
−K(t2p+1 −t2p ) ψ0 −K(t2p+1 −t2p )
max 1 − e ,1 − (1 − e ) = 1 − e−K(t2p+1 −t2p ) .
K

So, (2.9) becomes

+∞
X  
ln 1 − e−K(t2p+1 −t2p ) = −∞. (2.13)
p=0

However, the above condition (2.13) is automatically satised since we can assume that

t2n+2 − t2n+1 ≤ T, ∀n ∈ N0 ,

for some T > 0, eventually splitting the intervals of positive interaction into subintervals of
length at most T .

3 Preliminaries
Let {xi }i=1,...,N be solution to (2.1) under the initial conditions (2.2). In this section, we present
some preliminary lemmas. We assume that the hypotheses of Theorem 2.1 are satised. We rst
give some results that are related to the behaviour of the solution {xi }i=1,...,N in the intervals
having a positive interaction.

4
Lemma 3.1. For each v ∈ IRd and n ∈ N0 , we have that
min hxj (t2n ), vi ≤ hxi (t), vi ≤ max hxj (t2n ), vi, (3.1)
j=1,...,N j=1,...,N

for all t ∈ [t2n , t2n+1 ] and i = 1, . . . , N .

Proof. Let n ∈ N0 . Given a vector v ∈ IRd , we set
Mt2n = max hxj (t2n ), vi.
j=1,...,N

For all  > 0, let us dene

 

K := t ∈ [t2n , t2n+1 ] : max hxi (s), vi < Mt2n + , ∀s ∈ [t2n , t) .
i=1,...,N

Denoted with
S  := sup K  ,
by continuity it holds that S  ∈ (t2n , t2n+1 ].
We claim that S  = t2n+1 . Indeed, suppose by contradiction that S  < t2n+1 . Note that by
denition of S  it turns out that

max hxi (t), vi < Mt2n + , ∀t ∈ (t2n , S  ), (3.2)

i=1,...,N

and
lim max hxi (t), vi = Mt2n + . (3.3)
t→S − i=1,...,N

For all i = 1, . . . , N and t ∈ (t2n , S  ), we compute

d 1 X
hxi (t), vi = α(t)ψ(xi (t), xj (t))hxj (t) − xi (t), vi.
dt N −1
j:j6=i

Notice that, being t ∈ (t2n , S  ), then t ∈ (t2n , t2n+1 ) and, as a consequence, it holds that α(t) = 1.
Thus, using (2.4) and (3.2), we can write
d 1 X
hxi (t), vi ≤ ψ(xi (t), xj (t))(Mt2n +  − hxi (t), vi)
dt N −1
j:j6=i

≤ K(Mt2n +  − hxi (t), vi).

Then, from Gronwall's inequality, we get
Z t
−K(t−t2n )
hxi (t), vi ≤ e hxi (t2n ), vi + K(Mt2n + ) e−K(t−s) ds
t2n
−K(t−t2n ) −Kt
=e hxi (t2n ), vi + (Mt2n + )e (eKt − eKt2n )

= e−K(t−t2n ) hxi (t2n ), vi + (Mt2n + )(1 − e−K(t−t2n ) )

≤ e−K(t−t2n ) Mt2n + Mt2n +  − Mt2n e−K(t−t2n ) − e−K(t−t2n )

= Mt2n +  − e−K(t−t2n )
 −t )
= Mt2n +  − e−K(S 2n
,

5
for all t ∈ (t2n , S  ). We have so proved that, ∀i = 1, . . . , N,
 −t )
hxi (t), vi ≤ Mt2n +  − e−K(S 2n
, ∀t ∈ (t2n , S  ).

Thus, we get
 −t )
max hxi (t), vi ≤ Mt2n +  − e−K(S 2n
, ∀t ∈ (t2n , S  ). (3.4)
i=1,...,N

Letting t → S − in (3.4), from (3.3) we have that

 −t )
Mt2n +  ≤ Mt2n +  − e−K(S 2n
< Mt2n + ,

which is a contradiction. Thus, S  = t2n+1 . As a consequence, we have that

max hxi (t), vi < Mt2n + , ∀t ∈ (t2n , t2n+1 ).

i=1,...,N

From the arbitrariness of  we can conclude that

max hxi (t), vi ≤ Mt2n , ∀t ∈ [t2n , t2n+1 ],

i=1,...,N

from which
hxi (t), vi ≤ Mt2n , ∀t ∈ [t2n , t2n+1 ], ∀i = 1, . . . , N,
which proves the second inequality in (3.1).
Now, to prove the other inequality, let v ∈ IRd and dene

mt2n = min hxj (t2n ), vi.

j=1,...,N

Then, for all i = 1, . . . , N and t > t2n , by applying the second inequality in (3.1) to the vector
−v ∈ IRd we get
−hxi (t), vi = hxi (t), −vi ≤ max hxj (t2n ), −vi
j=1,...,N

= − min hxj (t2n ), vi = −mt2n ,

j=1,...,N

from which
hxj (s), vi ≥ mt2n .
Thus, also the rst inequality in (3.1) is fullled.

Lemma 3.2. For each n ∈ N0 and i, j = 1, . . . , N , we get

|xi (s) − xj (t)| ≤ d(t2n ), ∀s, t ∈ [t2n , t2n+1 ]. (3.5)

Proof. Fix n ∈ N0 and i, j = 1, . . . , N . Given s, t ∈ [t2n , t2n+1 ], if |xi (s) − xj (t)| = 0 then of
course d(t2n ) ≥ 0 = |xi (s) − xj (t)|. Thus, we can assume |xi (s) − xj (t)| > 0 and we set

xi (s) − xj (t)
v= .
|xi (s) − xj (t)|

6
It turns out that v is a unit vector and, using (3.1), we can write

|xi (s) − xj (t)| = hxi (s) − xj (t), vi = hxi (s), vi − hxj (t), vi

≤ max hxl (t2n ), vi − min hxl (t2n ), vi

l=1,...,N l=1,...,N

≤ max hxl (t2n ) − xk (t2n ), vi

l,k=1,...,N

≤ max |xl (t2n ) − xk (t2n )||v| = d(t2n ),

l,k=1,...,N

which proves (3.5).

Remark 3.3. Let us note that from (3.5), in particular, it follows that

d(t2n+1 ) ≤ d(t2n ), ∀n ∈ N0 . (3.6)

Indeed, let i, j = 1, . . . , N be such that

d(t2n+1 ) = |xi (t2n+1 ) − xj (t2n+1 )|.

Then, if d(t2n+1 ) = 0, of course we have d(t2n+1 ) = 0 ≤ d(t2n ). So we can assume d(t2n+1 ) > 0.
Then, we can apply (3.5) with s, t = t2n+1 and we get

d(t2n+1 ) = |xi (t2n+1 ) − xj (t2n+1 )| ≤ d(t2n ).

Now, we deal with the intervals in which the agents repeal each other.

Lemma 3.4. For each v ∈ IRd and n ∈ N0 , we have that

min hxj (t2n+2 ), vi ≤ hxi (t), vi ≤ max hxj (t2n+2 ), vi, (3.7)
j=1,...,N j=1,...,N

for all t ∈ [t2n+1 , t2n+2 ] and i = 1, . . . , N .

Proof. Let n ∈ N0 . Given a vector v ∈ IRd , we set
Mt2n+2 = max hxj (t2n+2 ), vi.
j=1,...,N

For all  > 0, let us dene

 

K := t ∈ [t2n+1 , t2n+2 ] : max hxi (s), vi < Mt2n+2 + , ∀s ∈ (t, t2n+2 ] .
i=1,...,N

Denoted with
S  := inf K  ,
by continuity it holds that S  ∈ [t2n+1 , t2n+2 ).
We claim that S  = t2n+1 . Indeed, suppose by contradiction that S  > t2n+1 . Note that by
denition of S  , it turns out that

max hxi (t), vi < Mt2n+2 + , ∀t ∈ (S  , t2n+2 ), (3.8)

i=1,...,N

7
and
lim max hxi (t), vi = Mt2n+2 + . (3.9)
t→S + i=1,...,N

For all i = 1, . . . , N and t ∈ (S  , t2n+2 ), we compute

d 1 X
hxi (t), vi = α(t)ψ(xi (t), xj (t))hxj (t) − xi (t), vi.
dt N −1
j:j6=i

Notice that, being t ∈ (S  , t2n+2 ), then t ∈ (t2n+1 , t2n+2 ) and, as a consequence, it holds that
α(t) = −1. Thus, using (2.4) and (3.8), we can write

d 1 X
hxi (t), vi ≥ − ψ(xi (t), xj (t))(Mt2n+2 +  − hxi (t), vi)
dt N −1
j:j6=i

≥ −K(Mt2n+2 +  − hxi (t), vi).

Then, from Gronwall's inequality, we get
Z t
K(t−S  ) 
hxi (t), vi ≥ e hxi (S ), vi − K(Mt2n+2 + ) eK(t−s) ds
S
 
= eK(t−S ) hxi (S  ), vi + (Mt2n+2 + )eKt (e−Kt − e−KS )
 
= eK(t−S ) hxi (S  ), vi + (Mt2n+2 + )(1 − eK(t−S ) ),

for all t ∈ (S  , t2n+2 ) and i = 1, . . . , N . Taking the supremum for i = 1, . . . , N , we get

) 
max hxi (t), vi ≥ eK(t−S max hxi (S  ), vi + (Mt2n+2 + )(1 − eK(t−S ) )
i=1,...,N i=1,...,N

 
= (Mt2n+2 + )eK(t−S ) + (Mt2n+2 + )(1 − eK(t−S ) ) = Mt2n+2 + .
For t = t2n+2 , we have that

Mt2n+2 = max hxi (t2n+2 ), vi ≥ Mt2n+2 +  > Mt2n+2 ,

i=1,...,N

which is a contradiction. Thus, S  = t2n+1 . As a consequence, we have that

max hxi (t), vi < Mt2n+2 + , ∀t ∈ (t2n+1 , t2n+2 ).

i=1,...,N

From the arbitrariness of  we can conclude that

max hxi (t), vi ≤ Mt2n+2 , ∀t ∈ [t2n+1 , t2n+2 ],

i=1,...,N

from which
hxi (t), vi ≤ Mt2n+2 , ∀t ∈ [t2n+1 , t2n+2 ], ∀i = 1, . . . , N.
Thus, the second inequality in (3.7) holds for all t ∈ [t2n+1 , t2n+2 ].
Now, to prove the other inequality, let v ∈ IRd and dene

mt2n+2 = min hxj (t2n+2 ), vi.

j=1,...,N

8
Then, for all i = 1, . . . , N and t ∈ [t2n+1 , t2n+2 ], by applying the second inequality in (3.7) to
the vector −v ∈ IRd we get

−hxi (t), vi = hxi (t), −vi ≤ max hxj (t2n+2 ), −vi

j=1,...,N

= − min hxj (t2n+2 ), vi = −mt2n+2 ,

j=1,...,N

from which
hxj (s), vi ≥ mt2n+2 .
Thus, also the rst inequality in (3.7) is fullled.

Lemma 3.5. For each n ∈ N0 and i, j = 1, . . . , N , we get

|xi (s) − xj (t)| ≤ d(t2n+2 ), ∀s, t ∈ [t2n+1 , t2n+2 ]. (3.10)

Proof. Fix n ∈ N0 and i, j = 1, . . . , N . Given s, t ∈ [t2n+1 , t2n+2 ], if |xi (s) − xj (t)| = 0 then of
course d(t2n+2 ) ≥ 0 = |xi (s) − xj (t)|. Thus, we can assume |xi (s) − xj (t)| > 0 and we set

xi (s) − xj (t)
v= .
|xi (s) − xj (t)|

It turns out that v is a unit vector and, using (3.1), we can write

|xi (s) − xj (t)| = hxi (s) − xj (t), vi = hxi (s), vi − hxj (t), vi

≤ max hxl (t2n+2 ), vi − min hxl (t2n+2 ), vi

l=1,...,N l=1,...,N

≤ max hxl (t2n+2 ) − xk (t2n+2 ), vi

l,k=1,...,N

≤ max |xl (t2n+2 ) − xk (t2n+2 )||v| = d(t2n+2 ),

l,k=1,...,N

which proves (3.10).

Remark 3.6. Let us note that from (3.5), in particular, it follows that

d(t2n+2 ) ≥ d(t2n+1 ), ∀n ∈ N0 . (3.11)

Indeed, let i, j = 1, . . . , N be such that

d(t2n+1 ) = |xi (t2n+1 ) − xj (t2n+1 )|.

Then, if d(t2n+1 ) = 0, of course we have d(t2n+2 ) = 0 ≤ d(t2n+1 ). So we can assume d(t2n+1 ) > 0.
Then, we can apply (3.10) with s, t = t2n+1 and we get

d(t2n+1 ) = |xi (t2n+1 ) − xj (t2n+1 )| ≤ d(t2n+2 ).

Now, with a quite simple argument, it can be proved that, in the intervals in which the
particles attract each other, the solutions of the system under consideration have a bound that
is uniform with respect to i = 1, . . . , N , but that depends of the maximum value assumed by the
opinions of the agents at the left end of the good interval.

9
Lemma 3.7. For every i = 1, . . . , N, we have that
0
|xi (t)| ≤ M2n , ∀t ∈ [t2n , t2n+1 ], (3.12)

where M2n
0 is that in (2.7).

Proof. Given i = 1, . . . , N and t ∈ [t2n , t2n+1 ], if |xi (t)| = 0 then trivially M2n
0 ≥ 0 = |x (t)|. On
i
the contrary, if |xi (t)| > 0, we dene
xi (t)
v= ,
|xi (t)|
which is a unit vector for which we can write

|xi (t)| = hxi (t), vi.

Then, using (3.1) and the Cauchy-Schwarz inequality we get

|xi (t)| ≤ max hxj (t2n ), vi ≤ max |xj (t2n )||v|

j=1,...,N j=1,...,N

0
= max |xj (t2n )| = M2n ,
j=1,...,N

which proves (3.12).

Now, we nd a bound from below for the inuence function ψ . This will be crucial for the
proof of the asymptotic consensus.

Proposition 3.8. Assume (2.12). Then, for all t ≥ 0, we have that

ψ(xi (t), xj (t)) ≥ ψ0 , ∀i, j = 1, . . . , N, (3.13)

where ψ0 is the positive constant in (2.10).

Remark 3.9. Let us note that the previous result holds in particular under assumption (2.8).
Indeed, we have already noticed that the condition (2.8) implies (2.12).

Proof of Proposition 3.8. From (3.12), it follows that

max |xi (t)| ≤ M 0 , ∀t ≥ 0. (3.14)
i=1,...,N

To see this, x t ≥ 0. Then, there exists n ∈ N0 such that t ∈ [t2n , t2n+2 ]. Thus, if t ∈ [t2n , t2n+1 ],
from (3.12) we have that
0
|xi (t)|≤ M2n = max |xi (t2n )|, ∀i = 1, . . . , N. (3.15)
i=1,...,N

On the other hand, assume that t ∈ (t2n+1 , t2n+2 ). Given i = 1, . . . , N , if |xi (t)|> 0, we dene
the unit vector
xi (t)
v= .
|xi (t)|
Then,
|xi (t)|= hxi (t), vi.

10
Now, for all s ∈ [t2n+1 , t), it holds that

d 1 X
hxi (s), vi = − ψ(xi (s), xj (s))(hxj (s), vi − hxi (s), vi)
ds N −1
j:j6=i

1 X
= ψ(xi (s), xj (s))(hxi (s), vi − hxj (s), vi).
N −1
j:j6=i

Thus, denoted with

mt2n+2 = min hxl (t2n+2 ), vi,
l=1,...,N

the rst inequality in (3.7) implies that

hxl (s), vi ≥ mt2n+2 , ∀s ∈ [t2n+1 , t], ∀l = 1, . . . , N.

As a consequence, we get
d 1 X
hxi (s), vi ≤ ψ(xi (s), xj (s))(hxi (s), vi − mt2n+2 )
ds N −1
j:j6=i

≤ K(hxi (s), vi − mt2n+2 ).

So, the Gronwall's inequality yields
Z s
hxi (s), vi ≤ eK(s−t2n+1 ) hxi (t2n+1 ), vi − Kmt2n+2 eK(s−r) dr
t2n+1

= eK(s−t2n+1 ) hxi (t2n+1 ), vi + mt2n+2 (1 − eK(s−t2n+1 ) ),

for all s ∈ [t2n+1 , t]. In particular, for s = t it comes that

hxi (t), vi ≤ eK(t−t2n+1 ) hxi (t2n+1 ), vi + mt2n+2 (1 − eK(t−t2n+1 ) )

= eK(t−t2n+1 ) (hxi (t2n+1 ), vi − mt2n+2 ) + mt2n+2

≤ eK(t2n+2 −t2n+1 ) (hxi (t2n+1 ), vi − mt2n+2 ) + mt2n+2

= eK(t2n+2 −t2n+1 ) hxi (t2n+1 ), vi + mt2n+2 (1 − eK(t2n+2 −t2n+1 ) )

≤ eK(t2n+2 −t2n+1 ) hxi (t2n+1 ), vi

≤ eK(t2n+2 −t2n+1 ) |xi (t2n+1 )|.

Thus, using (3.12) we get

|xi (t)| = hxi (t), vi ≤ eK(t2n+2 −t2n+1 ) M2n

0
.

Of course, the above inequality is satised also if |xi (t)| = 0. Thus, combining this last inequality
with (3.15), being eK(t2n+2 −t2n+1 ) > 1, we can conclude that

|xi (t)| ≤ eK(t2n+2 −t2n+1 ) M2n

0
, ∀n ∈ N0 , t ∈ [t2n , t2n+2 ], i = 1, . . . , N. (3.16)

11
Now, let us note that, using an induction argument, from (3.16) it follows that
n
Y
0
M2n+2 = max |xi (t2n+2 )| ≤ M00 eK(t2p+2 −t2p+1 ) , ∀n ≥ 0.
i=1,...,N
p=0

As a consequence, for all t ≥ 0, it holds

n
Y Pn
|xi (t)| ≤ M00 eK(t2p+2 −t2p+1 ) = eK p=0 (t2p+2 −t2p+1 ) M00 .
p=0

Then, for all t ≥ 0, P∞

|xi (t)| ≤ M00 eK p=0 (t2p+2 −t2p+1 ) ,
which proves (3.14).
Finally, from (3.14), we deduce that, for all t ≥ 0

ψ(xi (t), xi (t)) ≥ ψ0 .

Proposition 3.10. For all i, j = 1, . . . , N , unit vector v ∈ IRd and n ∈ N0 we have that
hxi (t) − xj (t), vi ≤ e−K(t−t̄) hxi (t̄) − xj (t̄), vi + (1 − e−K(t−t̄) )d(t2n ), (3.17)

for all t2n+1 ≥ t ≥ t̄ ≥ t2n .

Proof. Fix n ∈ N0 and v ∈ IRd such that |v| = 1. We set
Mt2n = max hxi (t2n ), vi,
i=1,...,N

mt2n = min hxi (t2n ), vi.

i=1,...,N

Then, it is easy to see that Mt2n − mt2n ≤ d(t2n ). Now, for all i = 1, . . . , N and t2n+1 > t ≥ t̄ ≥
t2n , from (3.1) we have that
d X
hxi (t), vi = α(t)aij (t)hxj (t) − xi (t), vi
dt
j:j6=i
1 X
= ψ(xi (t), xj (t))(hxj (t), vi − hxi (t), vi)
N −1
j:j6=i
1 X
≤ ψ(xi (t), xj (t))(Mt2n − hxi (t), vi).
N −1
j:j6=i

Note that hxi (t), vi ≤ Mt2n from (3.1), so that Mt2n − hxi (t), vi ≥ 0. Therefore, using (2.4), we
can write
d 1 X
hxi (t), vi ≤ K (Mt2n − hxi (t), vi) = K(Mt2n − hxi (t), vi).
dt N −1
j:j6=i

Thus, from the Gronwall's inequality it comes that

Z t
hxi (t), vi ≤ e−K(t−t̄) hxi (t̄), vi + KMt2n e−K(t−t̄)+K(s−t̄) ds
t̄

12
 = e−K(t−t̄) hxi (t̄), vi + e−K(t−t̄) Mt2n (eK(t−t̄) − 1),
that is
hxi (t), vi ≤ e−K(t−t̄) hxi (t̄), vi + (1 − e−K(t−t̄) )Mt2n . (3.18)
Now, for all i = 1, . . . , N and t2n+1 > t ≥ t̄ ≥ t2n , using (3.1) it holds

d 1 X
hxi (t), vi = α(t)ψ(xi (t), xj (t)(hxj (t), vi − hxi (t), vi)
dt N −1
i:j6=i
1 X
≥ ψ(xi (t), xj (t))(mt2n − hxi (t), vi).
N −1
j:j6=i

Then, since hxi (t), vi ≥ mt2n from (3.1) and by recalling that ψ is bounded from (2.4), we get

d
hxi (t), vi ≥ K(mt2n − hxi (t), vi).
dt
Thus, by using the Gronwall's inequality, it turns out that

hxi (t), vi ≥ e−K(t−t̄) hxi (t̄), vi + (1 − e−K(t−t̄) )mt2n . (3.19)

Therefore, for all i, j = 1, . . . , N and t2n+1 > t ≥ t̄ ≥ t2n , by using (3.18) and (3.19) and by
recalling that Mt2n − mt2n ≤ d(t2n ), we nally get

hxi (t) − xj (t), vi = hxi (t), vi − hxj (t), vi

≤ e−K(t−t̄) hxi (t̄), vi + (1 − e−K(t−t̄) )Mt2n

−e−K(t−t̄) hxj (t̄), vi − (1 − e−K(t−t̄) )mt2n

= e−K(t−t̄) hxi (t̄) − xj (t̄), vi + (1 − e−K(t−t̄) )(Mt2n − mt2n )

≤ e−K(t−t̄) hxi (t̄) − xj (t̄), vi + (1 − e−K(t−t̄) )d(t2n ),

i.e. (3.17) holds true.

Thanks to the presence of a uniform bound on the inuence function ψ , the following funda-
mental estimate holds in the intervals of positive interaction.

Proposition 3.11. For all n ∈ N0 , there exists a constant C2n ∈ (0, 1), independent of N ∈ N0 ,
such that
d(t2n+1 ) ≤ C2n d(t2n ). (3.20)

Proof. Let n ∈ N0 . Trivially, if d(t2n+1 ) = 0, then of course inequality (3.20) holds for any
positive constant. So, suppose d(t2n+1 ) > 0. Let i, j = 1, . . . , N be such that d(t2n+1 ) =
|xi (t2n+1 ) − xj (t2n+1 )|. We set

xi (t2n+1 ) − xj (t2n+1 )
v= .
|xi (t2n+1 ) − xj (t2n+1 )|

Then, v is a unit vector for which we can write

d(t2n+1 ) = hxi (t2n+1 ) − xj (t2n+1 ), vi.

13
Let us dene
Mt2n = max xl (t2n ), vi,
l=1,...,N

mt2n = min hxl (t2n ), vi.

l=1,...,N

Then Mt2n − mt2n ≤ d(t2n ).

Now, we distinguish two dierent situations.
Case I. Assume that there exists t̄ ∈ [t2n , t2n+1 ) such that
hxi (t̄) − xj (t̄), vi < 0.

Then, from (3.17) with t2n+1 ≥ t̄ ≥ t2n , we have

d(t2n+1 ) ≤ e−K(t2n+1 −t̄) hxi (t̄) − xj (t̄), vi + (1 − e−K(t2n+1 −t̄) )d(t2n )

≤ (1 − e−K(t2n+1 −t̄) )d(t2n ) (3.21)
≤ (1 − e−K(t2n+1 −t2n ) )d(t2n ).

Case II. Assume it rather holds

hxi (t) − xj (t), vi ≥ 0, ∀t ∈ [t2n , t2n+1 ). (3.22)

Then, for every t ∈ [t2n , t2n+1 ), we have that

d 1 X
hxi (t) − xj (t), vi = ψ(xi (t), xl (t))hxl (t) − xi (t), vi
dt N −1
l:l6=i
1 X
− ψ(xi (t), xl (t))hxl (t) − xj (t), vi
N −1
l:l6=j
1 X
= ψ(xi (t), xl (t))(hxl (t), vi − Mt2n + Mt2n − hxi (t), vi)
N −1
l:l6=i
1 X
+ ψ(xi (t), xl (t))(hxj (t), vi − mt2n + mt2n − hxl (t), vi)
N −1
l:l6=j
:= S1 + S2 .

Now, being t ∈ [t2n , t2n+1 ), from (3.1) we have that

mt2n ≤ hxk (t), vi ≤ Mt2n , ∀k = 1, . . . , N. (3.23)

Therefore, using (3.12), we get

1 X
S1 = ψ(xi (t), xl (t−))(hxl (t), vi − Mt2n )
N −1
l:l6=i
1 X
+ ψ(xi (t), xl (t))(Mt2n − hxi (t), vi)
N −1
l:l6=i
1 X
≤ ψ0 (hxl (t), vi − Mt2n ) + K(Mt2n − hxi (t), vi),
N −1
l:l6=i

14
and
1 X
S2 = ψ(xi (t), xl (t))(hxj (t), vi − mt2n )
N −1
l:l6=j
1 X
+ ψ(xi (t), xl (t))(mt2n − hxl (t), vi)
N −1
l:l6=j
1 X
≤ K(hxj (t), vi − mt2n ) + ψ0 (mt2n − hxl (t), vi).
N −1
l:l6=j

Combining this last fact with (3.23) it comes that

d
hxi (t) − xj (t), vi ≤ K(Mt2n − mt2n − hxi (t) − xj (t), vi)
dt
1 X
+ ψ0 (hxl (t), vi − Mt2n + mt2n − hxl (t), vi)
N −1
l:l6=i,j
1
+ ψ0 (hxj (t), vi − Mt2n + mt2n − hxi (t), vi)
N −1
N −2
= K(Mt2n − mt2n ) − Khxi (t) − xj (t), vi + ψ0 (−Mt2n + mt2n )
N −1
1
+ ψ0 (hxj (t), vi − Mt2n + mt2n − hxi (t2n ), vi).
N −1

Now, from (3.22) we get

d
hxi (t) − xj (t), vi ≤ K(Mt2n − mt2n ) − Khxi (t) − xj (t), vi
dt
N −2 1
+ ψ0 (−Mt2n + mt2n ) + ψ0 (−Mt2n + mt2n )
N −1 N −1
1
− ψ0 hxi (t) − xj (t), vi
N −1
≤ K(Mt2n − mt2n ) − Khxi (t) − xj (t), vi + ψ0 (−Mt2n + mt2n )

= (K − ψ0 ) (Mt2n − mt2n ) − Khxi (t) − xj (t), vi.

Hence, from Gronwall's inequality it comes that

hxi (t) − xj (t), vi ≤ e−K(t−t2n ) hxi (t2n ) − xj (t2n ), vi

Z t
+(Mt2n − mt2n ) (K − ψ0 ) e−K(t−s) ds,
t2n

15
for all t ∈ [t2n , t2n+1 ). In particular, for t = t2n+1 , from (3.5) it comes that
Z t2n+1
d(t2n+1 ) ≤ e−K(t2n+1 −t2n ) hxi (t2n ) − xj (t2n ), vi + (Mt2n − mt2n ) (K − ψ0 )e−K(t2n+1 −s) ds
t2n
Z t2n+1
≤ e−K(t2n+1 −t2n ) |xi (t2n ) − xj (t2n )| + (Mt2n − mt2n ) (K − ψ0 )e−K(t2n+1 −s) ds
t2n
 Z t2n+1 Z t2n+1 
−K(t2n+1 −t2n ) −K(t2n+1 −s) −K(t2n+1 −s)
≤ e +K e ds − ψ0 e ds d(t2n )
t2n t2n
 
−K(t2n+1 −t2n ) −K(t2n+1 −t2n ) ψ0 −K(t2n+1 −t2n )
= e +1−e − (1 − e ) d(t2n )
K
 
ψ0 −K(t2n+1 −t2n )
= 1− (1 − e ) d(t2n ).
K

So, if we set  
−K(t2n+1 −t2n ) ψ0 −K(t2n+1 −t2n )
C2n := max 1 − e ,1 − (1 − e ) , (3.24)
K
C2n ∈ (0, 1) is the constant for which (3.20) holds.

In the bad intervals, the previous estimate is not valid, being the diameter of the solution
evaluated at t2n+2 larger than the diameter evaluated at t2n+1 . However, the growth of the
diameter in the intervals with negative interaction can be in some way controlled, as the following
lemma shows.

Lemma 3.12. For all n ∈ N0 , we have that

eK(t2n+2 −t2n+1 )
d(t2n+2 ) ≤ d(t2n+1 ). (3.25)
2 − eK(t2n+2 −t2n+1 )
Proof. Let n ∈ N0 . Let all i, j = 1, . . . , N be such that
d(t2n+2 ) = |xi (t2n+2 ) − xj (t2n+2 )|.

If d(t2n+2 ) = 0, then from (3.11) also d(t2n+1 ) = 0 and of course inequality (3.25) is fullled. So,
we can assume that d(t2n+2 ) > 0. In this case, let v ∈ Rd be the so dened unit vector

xi (t2n+2 ) − xj (t2n+2 )
v= .
|xi (t2n+2 ) − xj (t2n+2 )|

Then,
d(t2n+2 ) = hxi (t2n+2 ) − xj (t2n+2 ), vi.
Moreover, we set
Mt2n+2 = max hxk (t2n+2 ), vi,
k=1,...,N

mt2n+2 = min hxk (t2n+2 ), vi.

k=1,...,N

Thus, from (3.7), for all t ∈ [t2n+1 , t2n+2 ], it holds

mt2n+2 ≤ hxk (t), vi ≤ Mt2n+2 , ∀k = 1, . . . , N. (3.26)

16
Now, for all t ∈ [t2n+1 , t2n+2 ), using the rst inequality in (3.26) and the fact that α(t) = −1,
we have that
d 1 X
hxi (t), vi = − ψ(xi (t), xl (t))(hxl (t), vi − hxi (t), vi)
dt N −1
l:l6=i

1 X
≤− ψ(xi (t), xl (t))(mt2n+2 − hxi (t), vi)
N −1
l:l6=i

1 X
= ψ(xi (t), xl (t))(hxi (t), vi) − mt2n+2 )
N −1
l:l6=i

≤ K(hxi (t), vi) − mt2n+2 ).

Therefore, the Gronwall's inequality yields

Z t
hxi (t), vi ≤ eK(t−t2n+1 ) hxi (t2n+1 ), vi − Kmt2n+2 eK(t−s) ds
t2n+1 (3.27)
K(t−t2n+1 ) K(t−t2n+1 )
=e hxi (t2n+1 ), vi + mt2n+2 (1 − e ).

On the other hand, using the second inequality in (3.26), we get

d 1 X
hxi (t), vi = − ψ(xj (t), xl (t))(hxl (t), vi − hxj (t), vi)
dt N −1
l:l6=j

1 X
≥− ψ(xj (t), xl (t))(Mt2n+2 − hxj (t), vi)
N −1
l:l6=j

1 X
= ψ(xj (t), xl (t))(hxj (t), vi) − Mt2n+2 )
N −1
l:l6=j

≥ K(hxj (t), vi) − Mt2n+2 ).

Hence, using the Gronwall's inequality, we can write

Z t
K(t−t2n+1 )
hxj (t), vi ≥ e hxj (t2n+1 ), vi − KMt2n+2 eK(t−s) ds
t2n+1 (3.28)
K(t−t2n+1 ) K(t−t2n+1 )
=e hxj (t2n+1 ), vi + Mt2n+2 (1 − e ).

Thus, combining (3.27) and (3.28), we can conclude that, for all t ∈ [t2n+1 , t2n+2 ], it holds

hxi (t) − xj (t), vi = hxi (t), vi − hxj (t), vi

≤ eK(t−t2n+1 ) hxi (t2n+1 ), vi + mt2n+2 (1 − eK(t−t2n+1 ) )

−eK(t−t2n+1 ) hxj (t2n+1 ), vi − Mt2n+2 (1 − eK(t−t2n+1 ) )

≤ eK(t−t2n+1 ) hxi (t2n+1 ) − xj (t2n+1 ), vi + (mt2n+2 − Mt2n+2 )(1 − eK(t−t2n+1 ) )

= eK(t−t2n+1 ) hxi (t2n+1 ) − xj (t2n+1 ), vi + (Mt2n+2 − mt2n+2 )(eK(t−t2n+1 ) − 1)

≤ eK(t−t2n+1 ) d(t2n+1 ) + d(t2n+2 )(eK(t−t2n+1 ) − 1).

17
For t = t2n+2 , we obtain
d(t2n+2 ) = hxi (t2n+2 ) − xj (t2n+2 ), vi ≤ eK(t2n+2 −t2n+1 ) d(t2n+1 ) + d(t2n+2 )(eK(t2n+2 −t2n+1 ) − 1),
from which
d(t2n+2 )(2 − eK(t2n+2 −t2n+1 ) ) ≤ eK(t2n+2 −t2n+1 ) d(t2n+1 ).
Thus, since 2 − eK(t2n+2 −t2n+1 ) > 0 from (2.6), we have that
eK(t2n+2 −t2n+1 )
d(t2n+2 ) ≤ d(t2n+1 ),
2 − eK(t2n+2 −t2n+1 )
i.e. (3.25) is proved.

4 Proof of Theorem 2.1

Proof of Theorem 2.1. Let {xi }i=1,...,N be solution to (2.1), (2.2). Then, for all n ∈ N0 , using
(3.20), (3.24) and (3.25)we have that
eK(t2n+2 −t2n+1 )
d(t2n+2 ) ≤ d(t2n+1 )
2 − eK(t2n+2 −t2n+1 )
eK(t2n+2 −t2n+1 )
 
−K(t2n+1 −t2n ) ψ0 −K(t2n+1 −t2n )
≤ max 1 − e ,1 − (1 − e ) d(t2n ).
2 − eK(t2n+2 −t2n+1 ) K
Thus, using an induction argument, we get
n !
eK(t2p+2 −t2p+1 )

Y ψ 0
d(t2n+2 ) ≤ K(t2p+2 −t2p+1 )
max 1 − e−K(t2p+1 −t2p ) , 1 − (1 − e−K(t2p+1 −t2p ) ) d(0)
2 − e K
p=0
 o
K(t2p+2 −t2p+1 ) n
ψ
max 1−e−K(t2p+1 −t2p ) ,1− K0 (1−e−K(t2p+1 −t2p ) )
P∞ e
p=0 ln K(t2p+2 −t2p+1 )
=e 2−e d(0)
   o
K(t2p+2 −t2p+1 )  n
ψ
+ln max 1−e−K(t2p+1 −t2p ) ,1− K0 (1−e−K(t2p+1 −t2p ) )
P∞ e
p=0 ln K(t2p+2 −t2p+1 )
=e 2−e d(0)
K(t2p+2 −t2p+1 )
P∞  
Now, p=0 ln e K(t2p+2 −t2p+1 ) < +∞ from (2.8). Then, the solution {xi }i=1,...,N converges to
2−e
consensus since the following condition is satised from (2.9):
∞   
X
−K(t2p+1 −t2p ) ψ0 −K(t2p+1 −t2p )
ln max 1 − e ,1 − (1 − e ) = −∞.
K
p=0

Under a stronger condition on the sequence {tn }n , the following exponential convergence to
consensus holds.
Theorem 4.1. Let ψ : IRd × IRd → IR be a positive, bounded, continuous function. Assume
that the sequence {tn }n of denition (2.5) satises (2.6). Assume (2.12) and that the following
condition holds:
!
eK(t2n+2 −t2n+1 )

ψ0
sup max 1 − e−K(t2n+1 −t2n ) , 1 − (1 − e−K(t2n+1 −t2n ) ) = c < 1.
n∈N 2 − eK(t2n+2 −t2n+1 ) K
(4.1)

18
Then, every solution {xi }i=1,...,N to (2.1) with the initial conditions (2.2) satises the following
exponential decay estimate
ln 2
d(t) ≤ e−γ (t− K
−T ) d(0), ∀t ≥ 0, (4.2)

for two suitable positive constants γ and T , independent of N .

Remark 4.2. The assumption (4.1) implies (2.9). Indeed, if (4.1) holds,
∞ !
eK(t2p+2 −t2p+1 )

X
−K(t2p+1 −t2p ) ψ0 −K(t2p+1 −t2p )
ln max 1 − e ,1 − (1 − e )
p=0
2 − eK(t2p+2 −t2p+1 ) K

∞
X
≤ ln c = −∞,
p=0

i.e.
∞ !
eK(t2p+2 −t2p+1 )

X ψ0
ln max 1 − e−K(t2p+1 −t2p ) , 1 − (1 − e−K(t2p+1 −t2p ) ) = −∞.
p=0
2 − eK(t2p+2 −t2p+1 ) K
(4.3)
eK(t2p+2 −t2p+1 )
Therefore, being > 1, it comes that
2−eK(t2p+2 −t2p+1 )

∞   
X
−K(t2p+1 −t2p ) ψ0 −K(t2p+1 −t2p )
ln max 1 − e ,1 − (1 − e )
K
p=0

∞ !
eK(t2p+2 −t2p+1 )

X ψ 0
≤ ln max 1 − e−K(t2p+1 −t2p ) , 1 − (1 − e−K(t2p+1 −t2p ) ) .
p=0
2 − eK(t2p+2 −t2p+1 ) K

Then, from (4.3), the condition (2.9) is satised.

Proof of Theorem 4.1. Let {xi }i=1,...,N be solution to (2.1), (2.2). Then, since (2.12) holds, for
all n ∈ N0 ,

eK(t2n+2 −t2n+1 )
 
−K(t2n+1 −t2n ) ψ0 −K(t2n+1 −t2n )
d(t2n+2 ) ≤ max 1 − e ,1 − (1 − e ) d(t2n ).
2 − eK(t2n+2 −t2n+1 ) K

Therefore, using (4.1), we get

d(t2n+2 ) ≤ cd(t2n ), (4.4)
with c ∈ (0, 1). As a consequence, using an induction argument,

d(t2n ) ≤ cn d(0), ∀n ∈ N0 . (4.5)

Now, let t ≥ 0. Then, there exists n ∈ N0 such that t ∈ [t2n , t2n+2 ). Thus, if t ∈ [t2n , t2n+1 ],
using (3.5),
d(t) ≤ d(t2n ).
On the other hand, if t ∈ (t2n+1 , t2n+2 ), from (3.10) and (4.4) we get

d(t) ≤ d(t2n+2 ) ≤ cd(t2n ).

19
Therefore, being c < 1, in both cases

d(t) ≤ d(t2n ).

So, using (4.5), we can write

1
d(t) ≤ cn d(0) = e−n ln( c ) d(0).

At this point, we distinguish two dierent situations.

Case I) Assume that
T := sup (t2n+1 − t2n ) < +∞.
n∈N0

So setting  
1 1
γ := ln ln 2
,
c K +T
it comes that
ln 2
d(t) ≤ e−nγ ( K
+T )

Now, using (2.6), it holds that t2n+2 ≤ (n+ 1) lnK2 + T . Thus, being t ≤ t2n+2 , we can conclude

that
ln 2
d(t) ≤ e−γ (t− K −T ) d(0),
which proves (4.2).
Case II) Assume that
sup(t2n+1 − t2n ) = +∞.
n∈N

We pick T̃ > 0. Without loss of generality, eventually splitting the intervals in which the weight
function α = 1 in subintervals of length at most T̃ , we can assume that

t2n+1 − t2n ≤ T̃ , ∀n ∈ N0 . (4.6)

Then, setting  
1 1
γ := ln ln 2
.
c K + T̃
Then, reasoning as in the previous case, we get that
ln 2
d(t) ≤ e−γ (t− K
−T̃ ) d(0),

which proves (4.2).

5 A particular case
In this section, we analyze a Hegselmann-Krause model of the type (2.1) with communication
rates aij of the form
1
aij (t) := ψ̃(|xi (t) − xj (t)|), (5.7)
N −1
where ψ̃ : IR → IR is a positive, continuous, and bounded function whose supremum norm is
given by
K̃ := kψ̃k∞ .

20
A class of functions that is included in this analysis is the one of all continuous, nonincreasing
functions ψ̃ satisfying
ψ̃(x) → 0, as x → +∞,
that is the class of functions more often appearing in the applications.
The following theorem guarantees the asymptotic consensus for the solution of the Hegselmann-
Krause type model (2.1) with the initial conditions (2.2) and communication rates given by
(5.7).

Theorem 5.1. Let ψ̃ : IR → IR be a positive, bounded, continuous function. Assume that the
sequence {tn }n of denition (2.5) satises
ln 2
t2n+2 − t2n+1 < , ∀n ∈ N0 . (5.8)
K̃
Moreover, assume that the two following conditions hold:
∞
!
X eK̃(t2p+2 −t2p+1 )
ln < +∞, (5.9)
p=0 2 − eK̃(t2p+2 −t2p+1 )

∞
( )!
X
−K̃(t2p+1 −t2p ) ψ̃0 −K̃(t2p+1 −t2p )
ln max 1 − e ,1 − (1 − e ) = −∞. (5.10)
p=0
K̃
where
ψ̃0 := min ψ̃(y), (5.11)
|y|≤M 0

being  
P∞ K̃(t2n+2 −t2n+1 )
e
p=0 ln
M̃ 0 := e 2−e
K̃(t2n+2 −t2n+1 )
d(0). (5.12)
Then, every solution {xi }i=1,...,N to (2.1) with the initial conditions (2.2) converges to consensus.
Proof. Using analogous arguments to the ones employed in the more general case examined in
the previous sections, all the results from Lemma 3.1 to Proposition 3.10 of Section 2 are still
valid, up to substituting the positive constant K with the constant of boundness of the now
considered inuence function K̃ . Therefore, for all n ∈ N0 , from (3.10) and (3.25) we have that

eK̃(t2n+2 −t2n+1 ) eK̃(t2n+2 −t2n+1 )

d(t2n+2 ) ≤ d(t2n+1 ) ≤ d(t2n ).
2 − eK̃(t2n+2 −t2n+1 ) 2 − eK̃(t2n+2 −t2n+1 )
Then,  
n
! K̃(t2n+2 −t2n+1 )
eK̃(t2n+2 −t2n+1 )
Pn e
p=0 ln
Y K̃(t2n+2 −t2n+1 )
d(t2n ) ≤ d(0) = e 2−e d(0)
p=0 2 − eK̃(t2n+2 −t2n+1 )
 
P∞ K̃(t2n+2 −t2n+1 )
e
p=0 ln
K̃(t2n+2 −t2n+1 )
≤e 2−e d(0).
We have so proven that
d(t2n ) ≤ M̃ 0 , ∀n ∈ N0 . (5.13)

21
As a consequence, for all t ≥ 0, since t ∈ [t2n , t2n+2 ), for some n ∈ N0 , using (3.5) and (3.10) it
comes that
d(t) ≤ M̃ 0 .
Thus, for all i, j = 1, . . . , N
|xi (t) − xj (t)| ≤ d(t) ≤ M̃ 0 ,
from which
ψ(|xi (t) − xj (t)|) ≥ ψ̃ 0 . (5.14)
Now, we repeat the proof of Lemma 3.11 up to substituting ψ0 with the constant ψ̃0 . To be
precise, we have that
( )
ψ̃0
d(t2n+1 ) ≤ max 1 − e−K̃(t2n+1 −t2n ) , 1 − (1 − e−K̃(t2n+1 −t2n ) ) d(t2n ),
K̃
n o
where max 1 − e−K̃(t2n+1 −t2n ) , 1 − ψ̃K̃0 (1 − e−K̃(t2n+1 −t2n ) ) ∈ (0, 1).
Then, taking into account of the conditions (5.9) and (5.10), the proof follows analogously to the
one of Theorem 2.1.

Again, with an additional condition on the sequence {tn }n , one can prove the following
exponential convergence to consensus result. We omit the proof since it is analogous to the one
of Theorem 4.1.
Theorem 5.2. Let ψ : IR → IR be a positive, bounded, continuous function. Assume that the
sequence {tn }n of denition (2.5) satises (5.8). Assume also that the following conditions hold:
+∞
X
(t2p+2 − t2p+1 ) < +∞, (5.15)
p=0
( )!
eK̃(t2n+2 −t2n+1 ) −K̃(t2n+1 −t2n ) ψ̃0
sup max 1 − e ,1 − (1 − e−K̃(t2n+1 −t2n ) ) = c < 1,
n∈N 2 − eK̃(t2n+2 −t2n+1 ) K̃
(5.16)
where ψ̃0 is the positive constant in (5.11). Then, every solution {xi }i=1,...,N to (2.1) with the
initial conditions (2.2) satises the following exponential decay estimate
 
−γ t− ln 2 −T
d(t) ≤ e K̃ d(0), ∀t ≥ 0, (5.17)

for two suitable positive constants γ , independent of N , and T .

6 The Cucker-Smale model with attractive-repulsive interaction

Consider a nite set of N ∈ N particles, with N ≥ 2. Let xi (t), vi (t) ∈ IRd be the opinion
and the velocity of the i-th particle at time t, respectively. The interactions between the ele-
ments of the system are described by the following Cucker-Smale model with attractive-repulsive
interaction

d
 dt xi (t) = vi (t), t > 0, ∀i = 1, . . . , N,
d P
α(t)aij (t)(xj (t) − xi (t)), t > 0, ∀i = 1, . . . , N, (6.18)
 dt vi (t) =
j:j6=i

22
with initial conditions (
xi (0) = x0i ∈ Rd , ∀i = 1, . . . , N,
(6.19)
vi (0) = vi0 ∈ Rd , ∀i = 1, . . . , N.

Here, the communication rates aij are of the form (5.7), being ψ̃ : IR → IR a continuous, positive,
bounded inuence function with supremum norm

K̃ = kψ̃k∞ .

The weight function α : [0, +∞) → {−1, 1} is dened as in (2.5). Moreover, we assume that the
sequence {tn }n in the denition (2.5) satises the following property:

1
t2n+1 − t2n > , ∀n ∈ N0 . (6.20)
K̃
Also, we set
M̃n0 := max |vi (tn )|. (6.21)
i=1,...,N

For each t ≥ 0, we dene the position diameter dX (·) as

dX (t) := max |xi (t) − xj (t)|,

i,j=1,...,N

and the velocity diameter dV (·) as

dV (t) := max |vi (t) − vj (t)|.

i,j=1,...,N

Denition 6.1. (Unconditional ocking) We say that a solution {(xi , vi )}i=1,...,N to system
(6.18) exhibits asymptotic ocking if it satises the two following conditions:

1. there exists a positive constant d∗ such that

sup dX (t) ≤ d∗ ;
t≥0

2. lim dV (t) = 0.
t→∞

We will prove the following ocking result.

Theorem 6.1. Let ψ̃ : IRd × IRd → IR be a positive, bounded, continuous function satisfying
Z ∞
min ψ̃(r)dx = +∞. (6.22)
0 r∈[0,x]

Assume that the sequence {tn }n of denition (2.5) satises (5.8), (6.20). Moreover, assume that
the condition (5.9) holds. Then, every solution {xi , vi }i=1,...,N to (6.18) with the initial conditions
(6.19) exhibits asymptotic ocking.

23
7 The asymptotic ocking result
Let {(xi , vi )}i=1,...,N be solution to (6.18) under the initial conditions (6.19). We assume that
the hypotheses of Theorem 6.1 are satised. We rst present some auxiliary lemmas that will
be needed for the proof of Theorem 6.1. Since these auxiliary results are analogous to the ones
given in Section 3, we omit their proof.
Lemma 7.1. For each v ∈ IRd and n ∈ N0 , we have that
min hvj (t2n ), vi ≤ hvi (t), vi ≤ max hvj (t2n ), vi, (7.1)
j=1,...,N j=1,...,N

for all t ∈ [t2n , t2n+1 ] and i = 1, . . . , N .

Lemma 7.2. For each n ∈ N0 and i, j = 1, . . . , N , we get
|vi (s) − vj (t)| ≤ dV (t2n ), ∀s, t ∈ [t2n , t2n+1 ]. (7.2)

Remark 7.3. Let us note that from (7.2), in particular, it follows that

dV (t2n+1 ) ≤ dV (t2n ), ∀n ∈ N0 . (7.3)

Lemma 7.4. For each v ∈ IRd and n ∈ N0 , we have that

min hvj (t2n+2 ), vi ≤ hvi (t), vi ≤ max hvj (t2n+2 ), vi, (7.4)
j=1,...,N j=1,...,N

for all t ∈ [t2n+1 , t2n+2 ] and i = 1, . . . , N .

Lemma 7.5. For each n ∈ N0 and i, j = 1, . . . , N , we get
|vi (s) − vj (t)| ≤ dV (t2n+2 ), ∀s, t ∈ [t2n+1 , t2n+2 ]. (7.5)

Remark 7.6. Let us note that from (7.2), in particular, it follows that

dV (t2n+2 ) ≥ dV (t2n+1 ), ∀n ∈ N0 . (7.6)

Lemma 7.7. For every i = 1, . . . , N, we have that

0
|vi (t)| ≤ M̃2n , ∀t ∈ [t2n , t2n+1 ], (7.7)

where M̃2n
0 is the positive constant in (6.21).

Lemma 7.8. For all t ≥ 0, we have that

max |vi (t)| ≤ M̃ 0 , (7.8)
i=1,...,N

where P∞
M̃ 0 := eK̃ p=0 (t2p+2 −t2p+1 ) M̃00 (7.9)
Proposition 7.9. For all i, j = 1, . . . , N , unit vector v ∈ IRd and n ∈ N0 we have that
hvi (t) − vj (t), vi ≤ e−K̃(t−t̄) hvi (t̄) − xj (t̄), vi + (1 − e−K̃(t−t̄) )dV (t2n ), (7.10)

for all t2n+1 > t ≥ t̄ ≥ t2n .

24
Lemma 7.10. For all n ∈ N0 , we have that
eK̃(t2n+2 −t2n+1 )
dV (t2n+2 ) ≤ dV (t2n+1 ). (7.11)
2 − eK̃(t2n+2 −t2n+1 )
Proposition 7.11. Assume (5.9). Then, for all t ≥ 0, it holds that
dV (t) ≤ M̄ 0 , (7.12)

where 
K̃(t2n+2 −t2n+1 )

P∞ e
p=0 ln
M̄ 0 := e 2−e
K̃(t2n+2 −t2n+1 )
dV (0). (7.13)

Proof. For all n ∈ N0 , from (7.3) and (7.11) we have that

eK̃(t2n+2 −t2n+1 ) eK̃(t2n+2 −t2n+1 )
dV (t2n+2 ) ≤ dV (t2n+1 ) ≤ dV (t2n ).
2 − eK̃(t2n+2 −t2n+1 ) 2 − eK̃(t2n+2 −t2n+1 )
Then,
 
n
! K̃(t2n+2 −t2n+1 )
eK̃(t2n+2 −t2n+1 )
Pn e
Y p=0 ln
K̃(t2n+2 −t2n+1 )
dV (t2n ) ≤ dV (0) = e 2−e dV (0)
p=0 2 − eK̃(t2n+2 −t2n+1 )
 
P∞ K̃(t2n+2 −t2n+1 )
e
p=0 ln
K̃(t2n+2 −t2n+1 )
≤e 2−e dV (0).
We have so proven that
dV (t2n ) ≤ M̄ 0 , ∀n ∈ N0 . (7.14)
As a consequence, for all t ≥ 0, since t ∈ [t2n , t2n+2 ), for some n ∈ N0 , using (3.5) and (3.10) we
can conclude that (7.12) holds true.

Now, we pick T > lnK̃2 . We can assume, eventually splitting the intervals of positive interac-
tion into subintervals of length at most T , that

t2n+1 − t2n ≤ T, ∀n ∈ N0 . (7.15)

As a consequence, from (5.8) and (7.15), being T > ln 2

K̃
, we can write

tn+1 − tn ≤ T, ∀n ∈ N0 . (7.16)

Denition 7.1. For all t ≥ 0, we dene

  
ψ̃t := min ψ̃(r) : r ∈ 0, max dX (s) , (7.17)
s∈[0,t]

and ( )
−K̃T e−K̃T
φ(t) := min e ψ̃t , . (7.18)
T

25
Remark 7.12. Let us note that, for all t ≥ 0 and i, j = 1, . . . , N ,

|xi (t) − xj (t)| ≤ max dX (s).

s∈[0,t]

As a consequence, it holds that

ψ̃(|xi (t) − xj (t)|) ≥ ψ̃t > 0, ∀t ≥ 0, i, j = 1, . . . , N. (7.19)

Proposition 7.13. For all n ∈ N0 ,

 Z t2n+1 
dV (t2n+1 ) ≤ 1− φ(s)ds dV (t2n ). (7.20)
t2n

Remark 7.14. Let us note that

Z t2n+1
φ(s)ds ∈ (0, 1), ∀n ∈ N0 ,
t2n

so that Z t2n+1
1− φ(s)ds ∈ (0, 1).
t2n

Proof of Proposition 7.13. Let n ∈ N0 . Trivially, if dV (t2n+1 ) = 0, then of course inequal-

ity (7.20) holds. So, suppose dV (t2n+1 ) > 0. Let i, j = 1, . . . , N be such that dV (t2n+1 ) =
|vi (t2n+1 ) − vj (t2n+1 )|. We set

vi (t2n+1 ) − vj (t2n+1 )
v= .
|vi (t2n+1 ) − vj (t2n+1 )|
Then, v is a unit vector for which we can write

dV (t2n+1 ) = hvi (t2n+1 ) − vj (t2n+1 ), vi.

Let us dene
Mt2n = max vl (t2n ), vi,
l=1,...,N

mt2n = min hvl (t2n ), vi.

l=1,...,N

Then Mt2n − mt2n ≤ dV (t2n ).

Now, we distinguish two dierent situations.
Case I. Assume that there exists t̄ ∈ [t2n , t2n+1 ] such that
hvi (t̄) − vj (t̄), vi < 0.

Then, from (7.10) with t2n+1 ≥ t̄ ≥ t2n , we have

dV (t2n+1 ) ≤ e−K̃(t2n+1 −t̄) hvi (t̄) − vj (t̄), vi + (1 − e−K̃(t2n+1 −t̄) )dV (t2n )
≤ (1 − e−K̃(t2n+1 −t̄) )dV (t2n )
≤ (1 − e−K̃T )dV (t2n ) (7.21)
 Z t2n+1 
≤ 1− φ(s)ds dV (t2n ).
t2n

26
 Case II. Assume it rather holds
hvi (t) − vj (t), vi ≥ 0, ∀t ∈ [t2n , t2n+1 ]. (7.22)
Then, for every t ∈ [t2n , t2n+1 ], we have that
d 1 X
hvi (t) − vj (t), vi = ψ̃(|xi (t) − xj (t)|)hvl (t) − vi (t), vi
dt N −1
l:l6=i
1 X
− ψ̃(|xi (t) − xj (t)|)hvl (t) − vj (t), vi
N −1
l:l6=j
1 X
= ψ̃(|xi (t) − xj (t)|)(hvl (t), vi − Mt2n + Mt2n − hvi (t), vi)
N −1
l:l6=i
1 X
+ ψ̃(|xi (t) − xj (t)|)(hvj (t), vi − mt2n + mt2n − hvl (t), vi)
N −1
l:l6=j
:= S1 + S2 .
Now, being t ∈ [t2n , t2n+1 ], from (7.1) we have that
mt2n ≤ hvk (t), vi ≤ Mt2n , ∀k = 1, . . . , N. (7.23)
Therefore, we get
1 X
S1 = ψ̃(|xi (t) − xj (t)|)(hvl (t), vi − Mt2n )
N −1
l:l6=i
1 X
+ ψ̃(|xi (t) − xj (t)|)(Mt2n − hvi (t), vi)
N −1
l:l6=i
1 X
≤ ψ̃t (hvl (t), vi − Mt2n ) + K̃(Mt2n − hvi (t), vi),
N −1
l:l6=i

and
1 X
S2 = ψ̃(|xi (t) − xj (t)|)(hvj (t), vi − mt2n )
N −1
l:l6=j
1 X
+ ψ̃(|xi (t) − xj (t)|)(mt2n − hvl (t), vi)
N −1
l:l6=j
1 X
≤ K̃(hvj (t), vi − mt2n ) + ψ̃t (mt2n − hvl (t), vi).
N −1
l:l6=j

Combining this last fact with (7.23) it comes that

d
hvi (t) − vj (t), vi ≤ K̃(Mt2n − mt2n − hvi (t) − vj (t), vi)
dt
1 X
+ ψ̃t (hvl (t), vi − Mt2n + mt2n − hvl (t), vi)
N −1
l:l6=i,j
1
+ ψ̃t (hvj (t), vi − Mt2n + mt2n − hvi (t), vi)
N −1
N −2
=˜
˜K(Mt2n − mt2n ) − K̃hvi (t) − vj (t), vi + ψ̃t (−Mt2n + mt2n )
N −1
1
+ ψ̃t (hvj (t), vi − Mt2n + mt2n − hvi (t2n ), vi).
N −1

27
Now, from (7.22) we get

d
hvi (t) − vj (t), vi ≤ K̃(Mt2n − mt2n ) − K̃hvi (t) − vj (t), vi
dt
N −2 1
+ ψ̃t (−Mt2n + mt2n ) + ψ̃t (−Mt2n + mt2n )
N −1 N −1
1
− ψ̃t hvi (t) − vj (t), vi
N −1

≤ K̃(Mt2n − mt2n ) − K̃hvi (t) − vj (t), vi + ψ̃t (−Mt2n + mt2n )

 
= K̃ − ψ̃t (Mt2n − mt2n ) − K̃hvi (t) − vj (t), vi.

Hence, from Gronwall's inequality it comes that

hvi (t) − vj (t), vi ≤ e−K̃(t−t2n ) hvi (t2n ) − vj (t2n ), vi

Z t  
+(Mt2n − mt2n ) K̃ − ψ̃s e−K̃(t−s) ds,
t2n

for all t ∈ [t2n , t2n+1 ]. In particular, for t = t2n+1 , from (7.2) it comes that
Z t2n+1
−K̃(t2n+1 −t2n )
dV (t2n+1 ) ≤ e hvi (t2n ) − vj (t2n ), vi + (Mt2n − mt2n ) (K̃ − ψ̃s )e−K̃(t2n+1 −s) ds
t2n
Z t2n+1
≤ e−K̃(t2n+1 −t2n ) |vi (t2n ) − vj (t2n )| + (Mt2n − mt2n ) (K̃ − ψ̃s )e−K̃(t2n+1 −s) ds
t2n
 Z t2n+1 Z t2n+1 
−K̃(t2n+1 −t2n ) −K̃(t2n+1 −s) −K̃(t2n+1 −s)
≤ e + K̃ e ds − ψ̃s e ds dV (t2n )
t2n t2n
 Z t2n+1 
−K̃(t2n+1 −t2n ) −K̃(t2n+1 −t2n ) −K̃(t2n+1 −s)
= e +1−e − ψ̃s e ds dV (t2n )
t2n
 Z t2n+1 
−K̃(t2n+1 −s)
= 1− ψ̃s e ds dV (t2n )
t2n
 Z t2n+1 
−K̃T
≤ 1−e ψ̃s ds dV (t2n )
t2n
 Z t2n+1 
≤ 1− φ(s)ds dV (t2n ).
t2n

So, taking into account (7.21), we can conclude that (7.20) holds.

Proof of Theorem 6.1. Let {(xi , vi )}i=1,...,N be solution to (6.18) under the initial conditions
(6.19). We dene the function D : [0, ∞) → [0, ∞),


dV (0),  t = 0,
 Rt
D(t) := 1 − t2n φ(s)ds dV (t2n ), t ∈ (t2n , t2n+1 ], n ∈ N0 ,
  
 1− t R
t2n+1 φ(s)ds dV (t2n+2 ), t ∈ (t2n+1 , t2n+2 ], n ∈ N0 .


28
By construction, D is piecewise continuous. Indeed, D is continuous everywhere except at points
tn , n ∈ N0 . Moreover, for all n ∈ N, n ≥ 1,
!
Z t2n
lim D(t) = dV (t2n ) ≥ 1− φ(s)ds dV (t2n ) = D(t2n ), (7.24)
t→t+
2n t2n−1

eK̃(t2n+2 −t2n+1 )
lim D(t) = dV (t2n+2 ) ≤ dV (t2n+1 )
t→t+
2n+1 2 − eK̃(t2n+2 −t2n+1 )
eK̃(t2n+2 −t2n+1 )
 Z t2n+1 
≤ 1− φ(s)ds dV (t2n ) (7.25)
2 − eK̃(t2n+2 −t2n+1 ) t2n
eK̃(t2n+2 −t2n+1 )
= D(t2n+1 ).
2 − eK̃(t2n+2 −t2n+1 )
Also, D is nonincreasing in all intervals of the form (tn , tn+1 ], n ∈ N0 .
Now, notice that, for almost all times t,

d d
max dX (s) ≤ dX (t) ,
dt s∈[0,t] dt

since maxs∈[0,t] dX (s) is constant or increases like dX (t). Moreover, for almost all times

d
dX (t) ≤ dV (t).
dt
Therefore, for almost all times

d d
max dX (s) ≤ dX (t) ≤ dV (t). (7.26)
dt s∈[0,t] dt

Next, we dene the function L : [0, ∞) → [0, ∞) as follows:

( )
max dX (s)
e−K̃T
Z
min e−K̃T
s∈[0,t]
L(t) := D(t) + min ψ̃(σ), dr,
0 σ∈[0,r] T

for all t ≥ 0. By denition, L is piecewise continuous, i.e. L is continuous everywhere except at

points tn , n ∈ N0 .
In addition, for each n ∈ N and for all t ∈ (t2n , t2n+1 ), we have that
( )
d d e −K̃T d
L(t) = D(t) + min e−K̃T ψ̃t , max dX (s)
dt dt T dt s∈[0,t]

d d
= D(t) + φ(t) max dX (s),
dt dt s∈[0,t]
and from (7.26) we get
d d
L(t) ≤ D(t) + φ(t)dV (t)
dt dt
= −φ(t)dV (t2n ) + φ(t)dV (t)

= φ(t)(dV (t) − dV (t2n )).

29
Thus, since dV (t) ≤ dV (t2n ) from (7.2), we can deduce that
d
L(t) ≤ 0, ∀t ∈ (t2n , t2n+1 ).
dt
As a consequence, for all t2n < s < t ≤ t2n+1 , it comes that

L(t) ≤ L(s).

Letting s → t+
2n , we get
( )
max dX (s)
e−K̃T
Z
min e−K̃T
s∈[0,t2n ]
L(t) ≤ lim L(s) = lim D(s) + min ψ̃(σ), dr,
s→t+
2n s→t+
2n 0 σ∈[0,r] T

for all t ∈ (t2n , t2n+1 ]. Thus, using (2.6), (7.2) and (7.24), we can write
( )
e−K̃T
Z max dX (s)
s∈[0,t2n ] −K̃T
L(t) ≤ dV (t2n ) + min e min ψ̃(σ), dr
0 σ∈[0,r] T
( )
1
Z max dX (s)
e −K̃T
min e−K̃T min ψ̃(σ),
s∈[0,t2n ]
=  D(t2n ) + dr
R t2n
1 − t2n−1 φ(s)ds 0 σ∈[0,r] T
" ( ) #
e−K̃T
Z max dX (s)
1 s∈[0,t2n ] − K̃T
≤  D(t2n ) + min e min ψ̃(σ), dr
Rt
1 − 2n φ(s)ds 0 σ∈[0,r] T
t2n−1

1
≤ R t2n  L(t2n ),
1− t2n−1 φ(s)ds

for all t ∈ (t2n , t2n+1 ]. So,

1
L(t) ≤  R t2n  L(t2n ), ∀t ∈ [t2n , t2n+1 ]. (7.27)
1− t2n−1 φ(s)ds

On the other hand, for all t ∈ (t2n+1 , t2n+2 ), we have that

( )
d d −K̃T e−K̃T d
L(t) = D(t) + min e ψ̃t , max dX (s)
dt dt T dt s∈[0,t]

d d
= D(t) + φ(t) max dX (s),
dt dt s∈[0,t]
and from (7.26) we get
d d
L(t) ≤ D(t) + φ(t)dV (t)
dt dt
= −φ(t)dV (t2n+2 ) + φ(t)dV (t)

= φ(t)(dV (t) − dV (t2n+2 )).

Thus, since dV (t) ≤ dV (t2n+2 ) from (7.5), we can deduce that
d
L(t) ≤ 0, ∀t ∈ (t2n+1 , t2n+2 ).
dt

30
As a consequence, for all t2n+1 < s < t ≤ t2n+2 , it comes that

L(t) ≤ L(s).

Letting s → t+
2n+1 , we get
( )
max dX (s)
e−K̃T
Z
min e−K̃T
s∈[0,t2n+1 ]
L(t) ≤ lim L(s) = lim D(s) + min ψ̃(σ), dr,
s→t+
2n+1 s→t+
2n+1 0 σ∈[0,r] T

for all t ∈ (t2n+1 , t2n+2 ]. Thus, using (7.25), we can write

( )
eK̃(t2n+2 −t2n+1 ) max dX (s)
e−K̃T
Z
min e−K̃T
s∈[0,t2n+1 ]
L(t) ≤ D(t2n+1 ) + min ψ̃(σ), dr
2 − eK̃(t2n+2 −t2n+1 ) 0 σ∈[0,r] T

eK̃(t2n+2 −t2n+1 )
≤ L(t2n+1 ),
2 − eK̃(t2n+2 −t2n+1 )
for all t ∈ (t2n+1 , t2n+2 ]. Therefore,

eK̃(t2n+2 −t2n+1 )
L(t) ≤ L(t2n+1 ), ∀t ∈ [t2n+1 , t2n+2 ]. (7.28)
2 − eK̃(t2n+2 −t2n+1 )
Now, combining (7.27) and (7.28), it turns out that

eK̃(t2n+2 −t2n+1 ) 1
L(t2n+2 ) ≤   L(t2n ), ∀n ∈ N. (7.29)
eK̃(t2n+2 −t2n+1 )
R t2n
2− 1− t2n−1 φ(s)ds

Thus, thanks to an induction argument, from (7.29) it follows that

 
n K̃(t −t )
 e 1
Y 2p+2 2p+1
L(t2n+2 ) ≤    L(t2 ), (7.30)
2 − e K̃(t2p+2 −t2p+1 ) R t2p
p=1 1− φ(s)ds t2p−1

for all n ∈ N.
Now, let t ≥ t4 . Then, there exists n ≥ 2 such that t ∈ [t2n , t2n+2 ]. As a consequence, if
t ∈ [t2n , t2n+1 ], from (7.27) and (7.30) with n − 1 ≥ 1, we get

1 eK̃(t2n+2 −t2n+1 ) 1
L(t) ≤ R t2n L(t2n ) ≤ L(t2n )
R t2n
1− t2n−1 φ(s)ds 2 − eK̃(t2n+2 −t2n+1 ) 1 − t2n−1 φ(s)ds
 
K̃(t −t ) n−1 K̃(t −t )
e 2n+2 2n+1 1 Y
 e
2p+2 2p+1 1
≤ R t2n    L(t2 )
K̃(t −t ) K̃(t
2 − e 2n+2 2n+1 1 − t2n−1 φ(s)ds p=1 2 − e 2p+2 2p+1 1 − 2p φ(s)ds −t ) R t
t2p−1
 
n K̃(t2p+2 −t2p+1 )
 e 1
Y
=    L(t2 ).
2 − eK̃(t2p+2 −t2p+1 ) 1 − t2p φ(s)ds
R
p=1 t2p−1

31
On the other hand, if t ∈ [t2n+1 , t2n+2 ], from (7.27), (7.28) and (7.30) it comes that

eK̃(t2n+2 −t2n+1) eK̃(t2n+2 −t2n+1) 1

L(t) ≤ L(t2n+1 ) ≤ R t2n L(t2n )
2 − eK̃(t2n+2 −t2n+1) 2 − eK̃(t2n+2 −t2n+1) 1 − t2n−1 φ(s)ds
 
n K̃(t −t )
 e 1
Y 2p+2 2p+1
≤    L(t2 ).
2 − e K̃(t2p+2 −t2p+1 ) R t2p
p=1 1− φ(s)ds t2p−1

Thus, for all t ≥ t4 ,

 
n
K̃(t2p+2 −t2p+1 )
e Y 1
L(t) ≤     L(t2 )
2 − eK̃(t2p+2 −t2p+1 ) 1 − t2p φ(s)ds
R
p=1 t2p−1
   (7.31)
 
Pn K̃(t2p+2 −t2p+1 )
ln e 1
p=1 +ln R t2p

K̃(t2p+2 −t2p+1 )
2−e 1− t φ(s)ds
=e 2p−1 L(t2 ).
  !
P+∞ eK̃(t2n+2 −t2p+1 ) P+∞
Now, from (5.9), p=1 ln < +∞. Also, p=1 ln Rt
1
< +∞.
2−eK̃(t2p+2 −t2p+1 ) 1− t 2p φ(s)ds
2p−1
Indeed, from (7.16), it turns out that
t2p
e−K̃T
Z
φ(s)ds ≤ (t2p − t2p−1 ), ∀p ≥ 1.
t2p−1 T

Then, since (5.9) holds (which implies (5.15)), we have that t2p − t2p−1 → 0, as p → ∞, and we
can write ! !
1 1
ln R t2p ≤ ln −K̃T
1 − t2p−1 φ(s)ds 1 − e T (t!2p − t2p−1 )
e−K̃T e−K̃T
= − ln 1 − (t2p − t2p−1 ) ∼ (t2p − t2p−1 ).
T T
!
P+∞ P+∞
As a consequence, since p=1 (t2p −t2p−1 ) < +∞ from (5.15), it holds p=0 ln R t2p
1
<
1− t2p−1 φ(s)ds
+∞.
So, setting   
 
P∞ K̃(t2n+2 −t2p+1 )
ln e 1
p=0 +ln R t2p

K̃(t2p+2 −t2p+1 )
2−e 1− t φ(s)ds
C := e 2p−1 L(t2 ),
taking into account of (7.31), we can conclude that

L(t) ≤ C, ∀t ≥ t4 . (7.32)

As a consequence, for all t ≥ t4 , by denition of L,

( )
max dX (s)
e−K̃T
Z
min e−K̃T
s∈[0,t]
min ψ̃(σ), dr ≤ L(t) ≤ C.
0 σ∈[0,r] T

32
Letting t → ∞ in the above inequality, we nally get
( )
sup dX (s)
e−K̃T
Z
s∈[0,∞)
min e−K̃T min ψ̃(σ), dr ≤ C. (7.33)
0 σ∈[0,r] T

Then, since the function ψ̃ satises (6.22), from (7.33), there exists a positive constant d∗ such
that
sup dX (s) ≤ d∗ . (7.34)
s∈[0,∞)

Now, we dene ( )
∗ −K̃T e−K̃T
φ := min e ψ∗ , ,
T
where
ψ∗ = min ψ̃(r).
r∈[0,d∗ ]

Note that φ∗ > 0, being ψ̃ a positive function. Also, from (7.34), it comes that
  
ψ∗ ≤ min ψ̃(r) : r ∈ 0, max dX (s) = ψ̃t ,
s∈[0,t]

for all t ≥ 0. Thus, we get

φ∗ ≤ φ(t), ∀t ≥ 0.
As a consequence, for all n ∈ N0 ,
Z t2n+1
φ(s)ds ≥ φ∗ (t2n+1 − t2n ),
t2n

from which Z t2n+1

1− φ(s)ds ≤ 1 − φ∗ (t2n+1 − t2n ).
t2n

So, recalling of inequality (7.20), we can write

dV (t2n+1 ) ≤ (1 − φ∗ (t2n+1 − t2n )) dV (t2n ), ∀n ∈ N0 . (7.35)

Now, using (7.11) and (7.35), we have that

eK̃(t2n+2 −t2n+1 ) eK̃(t2n+2 −t2n+1 ) t2n+1

 Z 
dV (t2n+2 ) ≤ dV (t2n+1 ) ≤ 1− φ(s)ds dV (t2n )
2 − eK̃(t2n+2 −t2n+1 ) 2 − eK̃(t2n+2 −t2n+1 ) t2n

eK̃(t2n+2 −t2n+1 )
≤ (1 − φ∗ (t2n+1 − t2n )) dV (t2n ).
2− eK̃(t2n+2 −t2n+1 )
Thus, using an induction argument, we get
n
!
Y eK̃(t2p+2 −t2p+1 ) ∗
dV (t2n+2 ) ≤ (1 − φ (t2p+1 − t2p )) dV (0)
p=0 2 − eK̃(t2p+2 −t2p+1 )
 
P∞ K̃(t2p+2 −t2p+1 )
e
p=0 ln (1−φ∗ (t2p+1 −t2p ))
K̃(t2p+2 −t2p+1 )
=e 2−e dV (0)

33
    
P∞ K̃(t2p+2 −t2p+1 )
e
p=0 ln +ln(1−φ∗ (t2p+1 −t2p ))
K̃(t2p+2 −t2p+1 )
=e 2−e dV (0)
 
P∞ eK̃(t2p+2 −t2p+1 )
Now, p=0 ln < +∞ from (5.9). Then, the solution {xi , vi }i=1,...,N exhibits
2−eK̃(t2p+2 −t2p+1 )
asymptotic ocking if the following condition is satised:
∞
X
ln ((1 − φ∗ (t2p+1 − t2p ))) = −∞. (7.36)
p=0

However, the above condition is guaranteed since, from (6.20),

φ∗
1 − φ∗ (t2p+1 − t2p ) ≤ 1 − ∈ (0, 1), ∀p ∈ N0 .
K̃
Thus,
∞ ∞
φ∗
X   X
∗
ln ((1 − φ (t2p+1 − t2p ))) ≤ ln 1 − = −∞,
p=0 p=0
K̃

from which (7.36) is fullled.

Acknowledgements. We thank INdAM-GNAMPA and UNIVAQ for the support.

References
[1] G. Ajmone, Marsan, N. Bellomo, and A. Tosin, Complex systems and society, Modeling
and simulation. SpringerBriefs in Mathematics. Springer, New York, 2013. xii+90 pp.

[2] R. Almeida, E. Girejko, L. Machado, A.B. Malinowska, and N. Martins, Application of

predictive control to the Hegselmann-Krause model. Math. Methods Appl. Sci., 41 (2018),
91919202.

[3] A. Aydogdu, M. Caponigro, S. McQuade, B. Piccoli, N. Pouradier Duteil, F. Rossi, and E.

Trélat, Interaction network, state space and control in social dynamics. Active particles.
Vol. 1. Advances in theory, models, and applications, 99140. Birkhäuser/Springer, Cham,
2017.

[4] N. Bellomo, M. A. Herrero, and A. Tosin, On the dynamics of social conict: Looking for
the Black Swan, Kinet. Relat. Models, 6, (2013), 459479.

[5] B. Bonnet and E. Flayac, Consensus and ocking under communication failures for a class
of Cucker-Smale systems, Systems Control Lett., 152, (2021), 10pp.

[6] A. Borzì and S. Wongkaew. Modeling and control through leadership of a rened ocking
system. Math. Models Methods Appl. Sci., 25, (2015), 255282.

Distributed control of robotic networks: a mathemat-

[7] F. Bullo, J. Cortés, and S. Martínez,
ical approach to motion coordination algorithms. Princeton series in applied mathematics.
Princeton University Press, Princeton, 2009.

34
 [8] S. Camazine, J. L. Deneubourg, N.R. Franks, J. Sneyd, G. Theraulaz, and E. Bonabeau,
Self-Organization in Biological Systems, Princeton University Press, Princeton, NJ, 2001.
[9] F. Ceragioli and P. Frasca, Continuous and discontinuous opinion dynamics with bounded
condence, Nonlinear Anal. Real World Appl., 13, (2012), 12391251.

[10] Y.-P. Choi and J. Haskovec, Cucker-Smale model with normalized communication weights
and time delay, Kinet. Relat. Models, 10, (2017), 10111033.

[11] Y.-P. Choi and Z. Li, Emergent behavior of Cucker-Smale ocking particles with heteroge-
neous time delays, Appl. Math. Lett., 86, (2018), 4956.

[12] Y.-P. Choi, A. Paolucci, and C. Pignotti, Consensus of the Hegselmann-Krause opinion
formation model with time delay, Math. Methods Appl. Sci., 44, (2021), 45604579.

[13] E. Continelli Asymptotic ocking for the Cucker-Smale model with time variable time
delays, Preprint 2022, arXiv:2206.11770.

[14] E. Continelli and C. Pignotti, Consensus for the Hegselmann-Krause model with time vari-
able time delays, Math. Meth. Appl. Sci., published online (2023), DOI: 10.1002/mma.9599.

[15] E. Continelli and C. Pignotti, Convergence to consensus results for Hegselmann-Krause

type models with attractive-lacking interaction, Preprint 2023, arXiv:2306.07658.

[16] F. Cucker and S. Smale, Emergent behavior in ocks, IEEE Transactions on Automatic
Control, 52, (2007), 852862.
[17] J.-G. Dong, S.-Y. Ha, K. Doheon, and K. Jeongho, Time-delay eect on the ocking in
an ensemble of thermomechanical Cucker-Smale particles, J. Dierential Equations, 266,
(2019), 23732407.

[18] J. Haskovec, Direct proof of unconditional asymptotic consensus in the Hegselmann-Krause

model with transmission-type delay, Bull. Lond. Math. Soc., 53, (2021), pp. 13121323.

[19] J. Haskovec and I. Markou, Asymptotic ocking in the Cucker-Smale model with reaction-
type delays in the non-oscillatory regime, Kinet. Relat. Models, 13, (2020), 795813.

[20] R. Hegselmann and U. Krause, Opinion dynamics and bounded condence models, analysis,
and simulation, J. Artif. Soc. Soc. Simul., 5, (2002), 124.

[21] M.O. Jackson, Social and economic networks, Princeton University Press, Princeton, NJ,
2008. xiv+504.

[22] A. Jadbabaie, J. Lin, and A. S. Morse, Coordination of groups of mobile autonomous agents
using nearest neighbor rules, IEEE Trans. Automat. Control, 48, (2003), 988-1001.

[23] M. Lachowicz, H. Leszczy«ski, and K. Topolski, Self-organization with small range interac-
tions: equilibria and creation of bipolarity, Appl. Math. Comput., 343, (2019), 156166.

[24] Y. Liu and J. Wu, Flocking and asymptotic velocity of the Cucker-Smale model with
processing delay, J. Math. Anal. Appl., 415, (2014), 5361.

35
 [25] J. Lu, D. W. C. Ho, and J. Kurths, Consensus over directed static networks with arbitrary
nite communications delays, Phys. Rev. E, 80, (2009), 066121, 7 pp.

[26] G. A. Marsan, N. Bellomo, and M. Egidi, Towards a mathematical theory of complex

socio-economical systems by functional subsystems representation, Kinet. Relat. Models, 1,
(2008), 249278.

[27] A. Paolucci and C. Pignotti, Consensus strategies for a Hegselmann-Krause model with
leadership and time variable time delay, J. Dynam. Dierential Equations, published online
(2023), https://doi.org/10.1007/s10884-023-10276-0.

[28] B. Piccoli, F. Rossi, and E. Trélat. Control to ocking of the kinetic Cucker-Smale model.
SIAM J. Math. Anal., 47, (2015), 46854719.
[29] B. Piccoli, N. Pouradier Duteil, and E. Trélat, Sparse control of Hegselmann-Krause models:
Black hole and declustering, SIAM J. Control Optim., 57, (2019), 26282659.

[30] C. Pignotti and E. Trélat, Convergence to consensus of the general nite-dimensional

Cucker-Smale model with time-varying delays, Commun. Math. Sci., 16, (2018), 20532076.

[31] S. Y. Pilyugin and M. C. Campi, Opinion formation in voting processes under bounded
condence, Netw. Heterog. Media, 14, (2019), 617632.

[32] M. Rodriguez Cartabia, Cucker-Smale model with time delay, Discrete Contin. Dynam.
Systems, 42, (2022), 24092432.
[33] S. Wongkaew, M. Caponigro, and A. Borzì, On the control through leadership of the
Hegselmann-Krause opinion formation model. Math. Models Methods Appl. Sci., 25, (2015),
565585.

36

View publication stats