Journal of Biological Dynamics

ISSN: 1751-3758 (Print) 1751-3766 (Online) Journal homepage: https://www.tandfonline.com/loi/tjbd20

A note on persistence about structured population

models

Kazuki Kawachi

To cite this article: Kazuki Kawachi (2008) A note on persistence about structured population
models, Journal of Biological Dynamics, 2:4, 449-464, DOI: 10.1080/17513750802213581

To link to this article: https://doi.org/10.1080/17513750802213581

Copyright Taylor and Francis Group, LLC

Published online: 10 Oct 2008.

Submit your article to this journal

Article views: 325

View related articles

Citing articles: 3 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=tjbd20
Journal of Biological Dynamics
Vol. 2, No. 4, October 2008, 449–464

A note on persistence about structured population models

Kazuki Kawachi*
Graduate School of Mathematical Sciences, The University of Tokyo, Tokyo, Japan

(Recived 30 August 2007; final version received 17 May 2008 )

In this paper, we report some results on persistence in two structured population models: a chronic-
age-structured epidemic model and an age-duration-structured epidemic model. Regarding these models,
we observe that the system is uniformly strongly persistent, which means, roughly speaking, that the
proportion of infected subpopulation is bounded away from 0 and the bound does not depend on the
initial data after a sufficient long time, if the basic reproduction ratio is larger than one. We derive this by
adopting Thieme’s technique, which requires some conditions about positivity and compactness. Although
the compactness condition is rather difficult to show in general infinite-dimensional function spaces, we
can apply Fréchet–Kolmogorov L1 -compactness criteria to our models. The two examples that we study
illuminate a useful method to show persistence in structured population models.

Keywords: structured population model; persistence; compactness condition

AMS Subject Classfication: 92D30; 35B40; 46B50

1. Introduction

Persistence and permanence are considered as important concepts of dynamical systems and of
systems in ecology, epidemics, etc. These concepts concern the long-term survival of each compo-
nent of a system of interacting components, for example, of species in an ecological community,
genotypes, strategies in an evolutionary game, or polynucleotides competing for energy sources.
In epidemics, they are related to the question of why the disease is never eradicated after a suffi-
ciently long time. Persistence means the long-term survival of some or all components of a system,
whereas permanence addresses the limits of growth for some (or all) components of the system.
We refer to Hutson and Schmitt [5] and Thieme [12] for background information and references.
An important theorem concerning them was obtained by Thieme [12](or see [15]), which shows
that uniform weak persistence implies uniform strong persistence. This means in the words of
population dynamics that, if the size of a population, while it may come arbitrarily close to 0 every
now and then, always climbs back to a level that is eventually independent of the initial data, then
the size is bounded away from 0 and the bound is eventually independent of the initial data.
The theorem has been gradually used in structured population models [2,3,8,10]. However, the
number of systems proved to have the property of persistence (permanence) has been at present

*Email: [email protected]

ISSN 1751-3758 print/ISSN 1751-3766 online

© 2008 Taylor & Francis
DOI: 10.1080/17513750802213581
http://www.informaworld.com
450 K. Kawachi

far fewer than those proved to have the property of the local asymptotic stability (or unstability) of
an equilibrium. Its major reason is that the former property was mathematically established later
than the latter. Another reason is that it is difficult to find an attracting (or absorbing) set and to
show compactness conditions linked with persistence and permanence. However, we consider the
problems here on a finite domain for the structural variable which require an approach somewhat
different from those used so far – that is, the method of using Thieme’s technique [14] and
Fréchet–Kolmogorov L1 -compactness criteria. The method is fully presented in [7, Section 8].
This paper is structured as follows. Section 2 presents some ideas from persistence theory. We
define the concept of persistence rigorously in the words of dynamical systems and see what
theorems are useful for showing uniform strong persistence in structured population models. We
illuminate the method by proving uniform strong persistence in two models in the following
sections. In Section 3, we deal with a chronic-age-structured epidemic model which was first pro-
posed by Martcheva and Castillo-Chavez [9]. In Section 4, we discuss an age-duration-structured
epidemic model which was first proposed by Inaba [6]. Section 5 summarizes our discussions and
presents an open problem.

2. Mathematical tools used in this paper

Let X be a metric space with metric d. Let  : [0, ∞) × X → X be a semiflow on X, i.e.,

t ◦ s = t+s for all t, s ≥ 0, where t denotes the mapping (t, ·) : X → X. Let ρ : X →
[0, ∞) be a non-negative uniformly continuous functional on X. We assume that the composition
σ : = ρ ◦  is a continuous mapping from [0, ∞) × X → R.

DEFINITION 2.1
(i)  is called weakly ρ-persistent if lim supt→∞ σ (t, x) > 0 whenever ρ(x) > 0.
(ii)  is called strongly ρ-persistent if lim inft→∞ σ (t, x) > 0 whenever ρ(x) > 0.
(iii)  is called uniformly weakly ρ-persistent if there exists some ε > 0 such that lim
supt→∞ σ (t, x) > ε whenever ρ(x) > 0.
(iv)  is called uniformly strongly ρ-persistent if there exists some ε > 0 such that lim
inft→∞ σ (t, x) > ε whenever ρ(x) > 0.

If no misunderstanding about the functional ρ is possible, we use persistent rather than

ρ-persistent.

Example 2.2 Let us consider the prey–predator model proposed by Lotka and Volterra:

x  (t) = x(t){α − βy(t)},

y  (t) = −y(t){γ − δx(t)},
x(0) = x0 ≥ 0, y(0) = y0 ≥ 0,

where x(t) denotes the population density of the prey at time t, y(t) the population density of the
predator at time t, and α, β, γ , δ are strictly positive constants. This system induces a continuous
semiflow  on the state space := R+ × R+ , where R+ denotes the set of non-negative real
numbers.
For example, we define ρ1 as ρ1 (x, y) := x, then σ1 (t, (x0 , y0 )) := ρ1 ((t, (x0 , y0 )))
represents the population density of the preys at time t under the condition that the popula-
tion density of the prey and that of the predator at time 0 are x0 and y0 , respectively. We
can easily find that  is strongly ρ1 -persistent, that it is uniformly weakly ρ1 -persistent with
 Journal of Biological Dynamics 451

lim supt→∞ 1 (t, (x, y)) > γ /(2δ) whenever ρ1 (x, y) > 0 and that it is not uniformly strongly
ρ1 -persistent because for any ε ∈ (0, γ /δ) lim inft→∞ σ1 (t, (ε, β/α)) = ε.
On the other hand, let us define ρ2 as ρ2 (x, y) := y, then σ2 (t, (x0 , y0 )) := ρ2 ((t, (x0 , y0 )))
represents the population density of the predator at time t under the condition that the population
density of the prey and that of the predator at time 0 are x0 and y0 , respectively. Note that  is not
weakly ρ2 -persistent, because the predators eventually die out if there are no prey at time t = 0.

We consider the following compactness condition (C):

There exist some ε0 > 0 and a closed subset B of X with the following properties.
(a) If ρ(x) ≤ ε0 , then limt→∞ d(t (x), B) = 0, that is, for every ε > 0, there exists some t > 0
such that for every s > t some b ∈ B can be found with d((s, x), b) < ε.
(b) For everyε1 ∈ (0, ε0 ), the intersection B ∩ ρ −1 [ε1 , ε0 ] is compact.

Remark 1 If we take B as t0 ( ), where t0 is a non-negative constant, then condition (C)(a) is

satisfied by itself for any ε. The reason is as follows: take any x ∈ , then for t > t0 , we have
t (x) = t0 (t−t0 (x)) ∈ t0 ( ) = B, which gives rise to the fact that d(t (x), B) = 0 for any
t > t0 .

The next theorem is due to Thieme [13].

THEOREM 2.3 Let the compactness condition (C) hold and assume that σ (t, x) > 0 for all t ≥ 0
whenever ρ(x) > 0. Then uniform weak persistence implies uniform strong persistence.

The reader may easily find the above theorem’s proof in [13,15].
When we apply this theorem to some structured population models, the metric space (X, d) is
often taken as the finite product space of L1 (S) for some S ⊂ Rn , where Rn denotes Euclidean
n-space. Hence, it is necessary to have some compactness criteria in function spaces in order
to verify the compactness condition (C)(b). Fréchet–Kolmogorov’s Theorem generalised to Rn
[1, Theorem IV.8.21] is a useful criterion.

THEOREM 2.4 Let S be a subset of Rn , B the σ -algebra of Borel subsets of S, μ the Lebesgue
measure on B, and 1 ≤ p < ∞. Then a subset K of Lp (S, B, μ) is relatively compact if and only
if it is bounded in Lp (S, B, μ) and the following limits are uniform for f ∈ K:

lim |f (x1 + h1 , . . . , xn + hn ) − f (x1 , . . . , xn )|p dx1 · · · dxn = 0, (1)
h1 ,...,hn →0 S

lim |f (x)|p dx = 0, (2)
M→∞ S\C
M

where CM is the cube −M ≤ x1 , . . . , xn ≤ M.

In the following, the notation of B and μ is omitted. If S is an interval of R and K is a subset

of L1 (S) consisting of absolutely continuous functions, condition (1) can be substituted into that
of uniform boundedness of the derivative of the functions in K, which is easier to prove. We will
use the following corollary in the next section.

COROLLARY 2.5 Let S be an interval of R and K a subset of L1 (S) consisting of absolutely

continuous functions. Then K is relatively compact if condition (2) holds and K and {f  |f ∈ K}
are bounded in L1 (S).
452 K. Kawachi

Proof We have to consider only condition (1), since the other conditions hold. From our
assumption, there exists a positive constant C such that f  L1 (S) ≤ C holds for any f ∈ K.
Then we have for x ∈ S and h > 0
  
 x+h  x+h
|f (x + h) − f (x)| =  f  (y) dy  ≤ |f  (y)| dy.
x x

By changing the order of integrals, we obtain

 
|f (x + h) − f (x)| ≤ |f  (y)|h dy = f  L1 (S) h ≤ Ch.
S S

The case h < 0 is the same. Hence we find that condition (1) holds and the limit is uniform for
f ∈ K. 

3. Persistence in a chronic-age-structured population model

Let us consider the following system proposed by Martcheva and Castillo-Chavez [9 and equa-
tions (3.2), (3.4), (3.5)] as an epidemiological model of hepatitis C with a chronic infectious class
structured by age-since-infection:

s  = b(1 − s) − γ is − sDu (t) + Au (t),

i  = γ is + sDu (t) − (b + k)i,
(∂t + ∂θ )u(θ, t) = −(b + α(θ ))u(θ, t), (3)
u(0, t) = ki,
s(0) = s0 ; i(0) = i0 ; u(θ, 0) = u0 (θ ),

with the relation

 ∞
s(t) + i(t) + u(θ, t)dθ = 1,
0

where
 ∞  ∞
Du (t) := δ(θ )u(θ, t) dθ, Au (t) := α(θ )u(θ, t) dθ.
0 0

s(t) denotes the portion of the susceptible population in the total population at time t, i(t) the
portion of the population infected with acute hepatitis C in the total population at time t and
u(θ, t) the age-since-infection density at time t – normalized by dividing by the total popu-
lation – of individuals infected with chronic hepatitis C (with or without cirrhosis), where θ
means the time spent in the chronic stage, b means birth/recruitment rate, γ the effective con-
tact rate of individuals with acute hepatitis C, k the rate of progression to chronic stage, and
α(θ ) the age-since-infection-structured treatment/recovery rate for the chronic stage, and δ(θ )
the age-since-infection-structured effective contact rate of individuals with chronic hepatitis C.
In the following, we change the notation of s, i, and u into x, y, and z, respectively, for con-
venience. Moreover, we introduce the finite maximum age-since-infection ω, which the original
paper [9] assumed infinite. The main reason for this modification is to simplify the discussion. In
reality, it is reasonable to assume that the maximum age-since-infection is finite, since the human
 Journal of Biological Dynamics 453

lifespan is finite. So, the system is rewritten as follows:

x  (t) = b{1 − x(t)} − γ x(t)y(t) − x(t)Dz (t) + Az (t),

y  (t) = γ x(t)y(t) + x(t)Dz (t) − (b + k)y(t),
(∂t + ∂θ )z(θ, t) = −{b + α(θ )}z(θ, t), (4)
z(0, t) = ky(t),
x(0) = x0 ; y(0) = y0 ; z(θ, 0) = z0 (θ )
with
 ω
x(t) + y(t) + z(θ, t) dθ = 1,
0
 ω  ω
Dz (t) := δ(θ )z(θ, t) dθ, Az (t) := α(θ )z(θ, t) dθ.
0 0

ASSUMPTION 3.1
(i) b, k, and γ are strictly positive constants.
(ii) δ(θ) is bounded above and strictly positive a.e. on [0, ω). We denote by δ ∞ the upper limit
of δ. ω
(iii) α(·) is non-negative, locally integrable on [0, ω) and 0 α(θ) dθ = ∞, which means that
every individual at z-stage returns into x-stage by the time the age-since-infection θ
reaches ω.

For this system, the basic reproduction ratio is defined as follows:

ω
γ + 0 δ(θ )kL(θ ) dθ
R0 := , (5)
b+k
θ
where L(θ ) := e−bθ exp(− 0 α(τ ) dτ ).
Martcheva and Castillo-Chavez [9] proved the following theorem, in which it makes no
difference whether ω is finite.

THEOREM 3.2 Let Assumption 3.1 hold, then we have:

(i) If R0 < 1, then the trivial equilibrium (x, y, z) = (1, 0, 0) is globally asymptotically
stable.
(ii) If R0 > 1, the trivial equilibrium is unstable
 ω and there is a unique non-trivial equilibrium,
which is locally asymptotically stable if 0 kL(θ )dθ < 1.

Next, let us consider the semiflow induced by system (4). We define its state space as follows:
   ω 

:= (x, y, z) ∈ R+ × R+ × L+ (0, ω)  x + y + z 1 = 1,
1
α(θ)z(θ) dθ < ∞ ,
0
ω
where f 1 := 0 |f (θ )| dθ for f ∈ L1 (0, ω) and

L1+ (0, ω) := {f ∈ L1 (0, ω) | f (θ) ≥ 0 a.e.}.

Then we can prove the following proposition. It means the system’s well-posedness, i.e., that
system (4) has a unique continuous solution with values in and the solution depends on its
initial value continuously.
454 K. Kawachi

PROPOSITION 3.3 System (4) induces a continuous semiflow  on the state space .

Proof From the third equation of system (4), we obtain


z0 (θ − t)e−bt (θ )/ (θ − t) if ω > θ > t,
z(θ, t) = (6)
ky(t − θ )L(θ ) if θ < min{t, ω},
θ
where (θ ) := exp(− 0 α(τ ) dτ ). We define α̃(θ ) := α(θ )L(θ ) and δ̃(θ) := δ(θ )L(θ ), and the
asterisk denotes the convolution operation defined by
 t
(f ∗ g)(t) := f (t − τ )g(τ ) dτ.
0

Substituting Equation (6) into the equation for x and y in system (4), we get

x  (t) = b{1 − x(t)} − γ x(t)y(t) − x(t){k(δ̃ ∗ y)(t) + F1 (t)} + k(α̃ ∗ y)(t) + F2 (t),
(7)
y  (t) = γ x(t)y(t) + x(t){k(δ̃ ∗ y)(t) + F1 (t)} − (b + k)y(t),

where we define α̃(θ ) = δ̃(θ ) = 0 for R \ [0, ω] and F1 (t), F2 (t) are given functions depending
on the initial data as
 ω
F1 (t) := δ(θ )z0 (θ − t)e−bt (θ)/ (θ − t) dθ,
t
 ω
F2 (t) := α(θ )z0 (θ − t)e−bt (θ)/ (θ − t) dθ
t

for t < ω and F1 (t) := 0, F2 (t) := 0 for t ≥ ω. Equations (7) are regarded as a two-dimensional
finitely retarded functional differential equation, since the minimum of retard is −ω. Then we can
obtain the conclusion by applying standard local existence and uniqueness results for retarded
functional differential equation [4, Section 2.2] to the integro-differential equation (7). Global
existence and continuous dependency on the initial value will follow immediately. 

Now, we come around to prepare for the proof of uniform strong persistence. Let

ρ(x, y, z) := y + z 1 , σ (t, (x, y, z)) := ρ((t, (x, y, z)))

for (x, y, z) ∈ and t ≥ 0. We can easily find that ρ : → R is non-negative and uniformly
continuous.
First, let us prove the positivity results which will be used for establishing uniform weak
persistence in Theorem 3.8.

LEMMA 3.4 We have z(θ, t) > 0 if either of the following two conditions hold:
(i) ω > θ > t and z0 (θ − t) > 0 or
(ii) θ < min{t, ω} and y(t − θ ) > 0.

Proof Let us see (6). Noticing that (s) > 0 for all 0 ≤ s < ω, we have the conclusion. 

LEMMA 3.5 If y0 > 0, then y(t) > 0 for all t ≥ 0.

 Journal of Biological Dynamics 455

Proof From the second equation of system (4) and the non-negativity of x, y, z, we have the
inequality
y  (t) ≥ −(b + k)y(t),
which leads to the inequality
y(t) ≥ y0 e−(b+k)t > 0. 

LEMMA 3.6 σ (t, (x0 , y0 , z0 )) > 0 for all t ≥ 0 whenever ρ(x0 , y0 , z0 ) > 0.

Proof If y0 > 0, then the assertion is obvious from Lemma 3.5. Let us consider the case y0 = 0
and z0 1 > 0.
Assume that, for some t0 > 0, σ (t0 , (x0 , y0 , z0 )) = 0 holds, i.e., y(t0 ) = 0 and z(θ, t0 ) = 0 for
a.e. θ ∈ [0, ω]. Then we can find from Lemma 3.4 (ii) that y(t) = 0 for all t ∈ [0, t0 ]. On the
other hand, the assumption of z0 1 > 0, Lemma 3.4, and the continuity of Dz (t) imply that there
exists some t1 > 0 such that Dz (t) > 0 for all t ∈ [0, t1 ). In addition, it is easily seen from the first
equation of system (4) that x(t) > 0 for all t > 0. Hence, x(t)Dz (t) > 0 for all t ∈ (0, t1 ), which
leads to y(t) > 0 for sufficiently small t > 0. This is contradictory. 

LEMMA 3.7 System (4) satisfies the compactness condition (C) with ε0 = 1 and B the closure
of T0 ( ), where T0 > ω is a sufficiently large positive number.

Proof Taking Remark 1 into consideration, we find that we have only to prove that B is relatively
compact in . Let P1 , P2 , and P3 be projections defined by
P1 (x, y, z) = x, P2 (x, y, z) = y, P3 (x, y, z) = z
for (x, y, z) ∈ R × R × L1 (0, ω). We have P1 B, P2 B ⊂ [0, 1], so P1 B and P2 B are relatively
compact. Then, all we have to do is to prove that P3 B is relatively compact, since the direct
product space of compact spaces is compact (Tychonoff’s theorem) and the direct product space
of the closure of topological spaces Xλ (λ ∈ ) is the closure of λ∈ Xλ . To this aim, we use
Corollary 2.5.
Let (x(t), y(t), z(·, t)) be a solution of system (4), where we take an initial condition
(x0 , y0 , z0 (·)) arbitrarily. Noticing that x(t), y(t), z(·, t) 1 ∈ [0, 1], it follows from the equation
for y in Equation (4) that
 ω 
 

|y (t)|  γ +  δ(θ )z(θ, t) dθ  + b + k
0
 ω
≤ γ + δ∞ z(θ, t) dθ + b + k
0

≤ γ + δ∞ + b + k (8)
for all t ≥ 0. And by the definition of L(θ ), we have that
 ω  ω
|L (θ )| dθ = − L (θ ) dθ
0 0

= L(0) − L(ω) = 1. (9)

By using Equations (6), (8) and (9), we obtain for any t ≥ T0
 ω
∂θ z(·, t) 1 = | − ky  (t − θ)L(θ ) + ky(t − θ)L (θ )| dθ
0

≤ k(γ + δ ∞ + b + k) L 1 + k,
456 K. Kawachi

which implies the boundedness of the set {z |(x, y, z(·)) ∈ B}. From this fact with the boundedness
of P3 B, we can apply Corollary 2.5 to derive that P3 B is relatively compact. This completes our
proof. 

And now let us prove the result on persistence with the above preparations, which imply that
system (4) satisfies all assumptions in Theorem 2.3. In the words of epidemiology, the rate of
individuals infected with acute or chronic hepatitis C is bounded away from 0 and the bound is
eventually independent of the initial data.

THEOREM 3.8 System (4) is uniformly strongly ρ-persistent if R0 > 1.

Proof Thanks to Theorem 2.3, we have to show only the uniform weak ρ-persistence for
system (4).
Let φd (t) := γ y(t + d) + Dz (t + d) for d ≥ 0. By applying the method of variation of constant
into Equations (4), we have
 t+d
−(b+k)(t+d)
y(t + d) = y0 e + x(τ ){γ y(τ ) + Dz (τ )}e−(b+k)(t+d−τ ) dτ
0
 t+d
≥ x(τ ){γ y(τ ) + Dz (τ )}e−(b+k)(t+d−τ ) dτ
d
 t
= φd (t − r)x(t + d − r)e−(b+k)r dr, (10)
0

where r := t + d − τ . Using this and Equation (6), for sufficiently large t, we have
 ω
Dz (t + d) = δ(θ )ky(t + d − θ )L(θ ) dθ
0
 ω  t+d−θ
≥ δ(θ )kL(θ ) x(τ ){γ y(τ ) + Dz (τ )}e−(b+k)(t+d−θ −τ ) dτ dθ
0 0
 ω  t+d
= δ(θ )kL(θ ) x(t + d − r){γ y(t + d − r)
0 θ

+ Dz (t + d − r)}e−(b+k)(r−θ) dr dθ
 t  r
= φd (t − r)x(t + d − r)e−(b+k)r δ(θ )kL(θ )e(b+k)θ dθ dr, (11)
0 0

where in the last line we changed the order of integrals and used the setting of δ(θ )L(θ ) = δ̃(θ):
δ̃(θ ) = 0 for R \ [0, ω]. We substitute Equations (10) and (11) into the expression of φd (t) to obtain
 t   r 
φd (t) ≥ φd (t − r)x(t + d − r)e−(b+k)r γ + δ(θ )kL(θ )e(b+k)θ dθ dr. (12)
0 0

Suppose that system (4) is not uniformly weakly ρ-persistent, then for any ε ∈ (0, 1) we can
choose some initial condition (x0 , y0 , z0 ) and some T > 0 which satisfy ρ(x0 , y0 , z0 ) > 0 and
 ω
y(t) + z(θ, t) dθ ≤ ε, x(t) ≥ 1 − ε, ∀t ≥ T .
0

If d > T is sufficiently large, then we have

 t   r 
φd (t) ≥ φd (t − r) · (1 − ε)e−(b+k)r γ + δ(θ )kL(θ )e(b+k)θ dθ dr
0 0
 Journal of Biological Dynamics 457

for all t ≥ 0. Taking Laplace transforms leads to

φ̂d (λ) ≥ φ̂d (λ)F (ε, λ), (13)

where    
∞ r
−λr −(b+k)r
F (ε, λ) := e (1 − ε)e γ+ δ(θ )kL(θ )e (b+k)θ
dθ dr.
0 0

Since φd is bounded, φ̂d (λ) is defined for all λ ≥ 0. We can easily find through the change of
integral variables that F (0, 0) = R0 > 1. Since F (ε, λ) is continuous, we see that F (ε, λ) > 1 if
ε, λ > 0 are chosen small enough. Then Equation (13) implies that φ̂d (λ) = 0, which means
φd (t) = 0 a.e. on [0, ∞), i.e., y(t) = Dz (t) = 0 for almost every t ≥ T . This contradicts
Lemma 3.6. 

4. Persistence in an age-duration-structured population model

Let us consider the following system proposed by Inaba [6] as an age-duration-structured

population model for HIV infection in a homosexual community:
(∂t + ∂a )x(t, a) = −λ(t, a)x(t, a),
(∂t + ∂τ )y(t, τ ; a) = 0,
x(t, 0) = 1, y(t, 0; a) = λ(t, a)x(t, a),
 ω b
λ(t, a) = C̃(P (t)) K(a, b, τ )y(t, τ ; b − τ ) dτ db, (14)
0 0
 ω   a 
P (t) = Bl(a) x(t, a) + (τ ; a − τ )y(t, τ ; a − τ ) dτ da,
0 0

x(0, a) = x0 (a), y(0, τ ; a) = y0 (τ ; a).

First, we shall follow [6] in order to derive Equation (14). We divide the homosexual population
into three groups: S (uninfected but susceptible), I (HIV infected) and A (fully developed AIDS
symptoms). Let S(t, a) be the age density of the S-population at time t and age a, I (t, τ ; a) the
density of the I -population at time t and age-since-infection (duration) τ which were infected
at age a, and A(t, τ ; a) the density of the A-population at time t and age-since-infection τ for
individuals who have developed AIDS at age a. Let B be the birth rate of the S-population, μ(a)
the age-specific natural death rate, γ (τ ; a) the rate of developing AIDS at disease age τ for the
S-individuals at age a, δ(τ ; a) the death rate at duration τ due to AIDS for the individuals who
have developed AIDS at age a and λ(t, a) the infection rate (the force of infection) at age a and
time t. Then, the dynamics of the population is governed by the following system:
(∂t + ∂a )S(t, a) = −{μ(a) + λ(t, a)}S(t, a),
(∂t + ∂τ )I (t, τ ; a) = −{μ(a + τ ) + γ (τ ; a)}I (t, τ ; a),
(∂t + ∂τ )A(t, τ ; a) = −{μ(a + τ ) + δ(τ ; a)}A(t, τ ; a),
S(t, 0) = B, I (t, 0; a) = λ(t, a)S(t, a), (15)
 a
A(t, 0; a) = γ (τ ; a − τ )I (t, τ ; a − τ ) dτ,
0

S(0, a) = S0 (a), I (0, τ ; a) = I0 (τ ; a), A(0, τ ; a) = A0 (τ ; a).

458 K. Kawachi

λ(t, a) is assumed to have the following expression:

 ω b
C(P (t))
λ(t, a) = β(a, b, τ ) I (t, τ ; b − τ ) dτ db,
0 0 P (t)
where β(a, b, τ ) is the transmission probability that an S-person of age a becomes infected by
sexual contact with an infected partner of age b and disease age τ, ω denotes the upper bound of
age of the sexually active population, P (t) is the total size of sexually active population given by
 ω  σ 
P (t) = S(t, σ ) + I (t, τ ; σ − τ ) dτ dσ
0 0

and C(P ) denotes the mean number of sexual partners an average individual has per unit time
when the population size is P .
Let us simplify system (15) by introducing new functions x, y by

S(t, a) = x(t, a)Bl(a), I (t, τ ; a) = i(t, τ ; a)Bl(a + τ )(τ ; a),

where l(a) and (a) are the survival functions defined by

a τ
l(a) := e− 0 μ(σ ) dσ
, (τ ; a) := e− 0 γ (σ ; a) dσ
.

Then, we obtain a simplified system (14) for (x, y), where

C(P )
C̃(P ) := , K(a, b, τ ) := β(a, b, τ )Bl(b)(τ ; b − τ ).
P
In the following, we make some assumptions for the coefficients and the functions appearing
in Equation (14), some of which can be satisfied when we impose appropriate restrictions on
μ, γ , β etc.

ASSUMPTION 4.1
(i) B and ω are strictly positive constants.
(ii) l : [0, ω] → R is a monotone nonincreasing Lebesgue-measurable function and l(0) = 1
and l(ω) = 0.
(iii)  : R+ 2
→ [0, 1] is Lebesgue-measurable and (· ; a) s monotone non-increasing for any
fixed a ∈ R+ .
(iv) K(a, b, τ ) is expressed as k1 (a)k2 (b, τ ), where
(a) k1 (a) > 0 if a ∈ [0, ω], and otherwise k1 (a) = 0. In addition, k1 ∈ L∞
+ (R) and we denote
its essential upper bound by k1∞ .
(b) If k2 (b, τ ) = 0, then (b, τ ) ∈ 0 := {(b, τ ) ∈ R2 | 0 ≤ τ ≤ b ≤ ω}. In addition, there
exists some constant a† ∈ (0, ω] such that k2 (b, τ ) > 0 whenever (b, τ ) ∈ D0 :=
{(b, τ ) ∈ 0 | b − τ ≤ a† }.
∞
(c) k2 ∈ L∞ 2
+ (R ) and we denote its essential upper bound by k2 .
(v) C̃(P ) is strictly positive, monotone non-increasing for P > 0 and there exists some constant
L > 0 such that for all P , Q > 0, we have

|C̃(P ) − C̃(Q)| ≤ L|P − Q|.

Let us generate additional ideas about Assumption 4.1 to supplement. (i) states that ω – the
upper bound of age of the sexually active population – is finite as in system (4) in Section 3,
which makes the following discussion easier; (iv) is the proportionate mixing assumption, that
 Journal of Biological Dynamics 459

is, there is no correlation between the age of susceptibles and the age of infectives, hence it is
not necessarily realistic but very helpful for theoretical analysis; (iv)(b) is a technical assumption,
stating that individuals who have infected by age a† have non-zero force of infection; (v) implies
that C0 := limP ↓0 C̃(P ) exists and is finite.
Mathematical well-posedness of the time evolution problem (14) is proved in [6, Appendix A],
where the semigroup solution can be constructed by using the perturbation method of non-densely
defined operators [11].
For system (14), the basic reproduction ratio is defined as follows:
 ω  ω
R0 := C̃(P0 ) k1 (z) k2 (b, b − z) db dz, (16)
0 z
ω
where P0 := 0 Bl(a) da.

LEMMA 4.2 The following relations hold for t > ω:

 a
x(t, a) + y(t, τ ; a − τ ) dτ = 1, (17)
0
 a
x(t, a) + (τ ; a − τ )y(t, τ ; a − τ ) dτ ≤ 1, (18)
0

P (t) ≤ P0 . (19)
C̃(P0 ) ≤ C̃(P (t)) ≤ C0 . (20)

Proof Let
 α
f (α) := x(t − a + α, α) + y(t − a + α, α − τ ; τ ) dτ,
0
then it is easy to show that f  (α) = 0, so f (α) does not depend on α and f (a) = f (0) =
x(t − a, 0) = 1. The other relations are obvious from Equations(14) and (17) and Assump-
tion 4.1. 

Let us consider the semiflow induced by this system. We define its state space := L1+ (0, ω) ×
L1+ (), where  := {(a, b) ∈ R2 | a ≥ 0, b ≥ 0, a + b ≤ ω}. Mathematical well-posedness of
the time evolution problem (14) implies the following proposition:

PROPOSITION 4.3 System (14) induces a continuous semiflow  on the state space .

Let
 ω  b
φ(t) := C̃(P (t)) k2 (b, τ ) y(t, τ ; b − τ ) dτ db.
0 0
Then we have λ(t, a) = k1 (a)φ(t) and for any t > τ

y(t, τ ; a) = y(t − τ, 0; a) = λ(t − τ, a)x(t − τ, a)

= k1 (a)φ(t − τ )x(t − τ, a). (21)

Let
 ω  b
ρ(x, y) := k2 (b, τ ) y(τ ; b − τ ) dτ db,
0 0

σ (t, (x, y)) := ρ((t, (x, y)))

460 K. Kawachi

for (x, y) ∈ and t ≥ 0. We can easily find that ρ : → R is non-negative and uniformly
continuous. In addition, we find that

φ(t) = C̃(P (t))σ (t, (x, y)). (22)

Then, let us prove the positivity results. First, the upper-boundedness of λ and the equation for
x in system (14) lead to
x(t, a) > 0 whenever t > a. (23)

LEMMA 4.4 If ρ(x0 , y0 ) > 0, then σ (t, (x0 , y0 )) > 0 for all t ≥ 0.

Proof Let (x(t, ·), y(t, · ; ·)) = (t, (x0 , y0 )). Suppose there exists some t0 > 0 such that
σ (t0 , (x0 , y0 )) = 0. Then, y(t0 , τ ; b − τ ) = 0 for a.e. (b, τ ) ∈ D0 .
If 0 ≤ τ ≤ b < t0 , from Equation (21), we have the relation

y(t0 , τ ; b − τ ) = k1 (b − τ )φ(t0 − τ )x(t0 − τ, b − τ ) (24)

and x(t0 − τ, b − τ ) > 0. Hence, we have φ(t0 − τ ) = 0 for almost every (b, τ ) satisfying 0 ≤
τ ≤ b < t0 and (b, τ ) ∈ D0 . From this and the continuity of φ, we have φ(0) = 0, and this and
the relation (22) give rise to ρ(x0 , y0 ) = σ (0, (x0 , y0 )) = 0, which contradicts the assumption
ρ(x0 , y0 ) > 0. 

Now, let us prove that system (14) satisfies the compactness condition (C). While we can apply
a multidimensional version of Corollary 2.5, it might be good to illuminate the technique of using
Theorem 2.4, which is slightly different from that in Section 3.

LEMMA 4.5 System (14) satisfies the compactness condition (C) with ε0 = 1 and B the closure
of T0 (), where T0 > 2ω is a sufficiently large positive number.

Proof Taking Remark 1 into consideration, we find that we have only to prove that B is relatively
compact in . Let P1 , P2 be projections defined by

P1 (x, y) = x, P2 (x, y) = y for (x, y) ∈ L1 (0, ω) × L1 ().

Our aim is to prove that P1 B and P2 B are relatively compact.

Let (x(t, ·), y(t, · ; ·)) be the solution of system (14), where we take an initial condition
(x0 (·), y0 (· ; ·)) arbitrarily. For convenience, we define x(t, a) = 0 for a ∈ R \ [0, ω] and
y(t, τ ; a) = 0 for (τ, a) ∈ R2 \ .
Since k1 ∈ L1 (R) and φ is uniformly continuous on [0, T0 ], which is based on its continuity
on [0, T0 ], for any ε > 0, there exists some δ = δ(ε) ∈ (0, ε) such that

|k1 (a + h) − k1 (a)| da < ε
R

for any h with |h| < δ and

|φ(t + h) − φ(t)| < ε

whenever h < δ and t, t + h ∈ [0, T0 ]. In the following, we fix such ε and δ.

 Journal of Biological Dynamics 461

Let 0 < a < ω − h and 0 < h < δ. Using Equation (14), Lemma 4.2 and
 ω  b
|φ(t)| ≤ C0 k2∞ y(t, τ, b − τ ) dτ db ≤ C0 k2∞ ω,
0 0

we can evaluate as follows:

|x(T0 , a + h) − x(T0 , a)|

  a+h a 
 
= e− 0 λ(T0 −a−h+r,r) dr − e− 0 λ(T0 −a+r,r) dr 
 a+h  a 
 
≤  λ(T0 − a − h + r, r) dr − λ(T0 − a + r, r) dr 
0 0
 a  h
≤ |λ(T0 − a + r, r + h) − λ(T0 − a + r, r)| dr + |λ(T0 − a − h + r, r)| dr
0 0
 a  h
= |k1 (r + h) − k1 (r)|φ(T0 − a + r) dr + k1 (r)φ(T0 − a − h + r) dr
0 0

≤ C0 k2∞ ω(1 + k1∞ )ε. (25)

Then we have
 ω
|x(T0 , a + h) − x(T0 , a)| da
0
 ω−h  ω
= |x(T0 , a + h) − x(T0 , a)| da + |x(T0 , a)| da
0 ω−h

≤ C0 k2∞ (1 + k1∞ )ω2 ε + h

≤ {C0 k2∞ (1 + k1∞ )ω2 + 1}ε.

The case h < 0 is the same. Hence, it follows from Theorem 2.4 that P1 B is relatively compact.
Next, let h1 , h2 ∈ (0, δ) and (τ, a), (τ + h1 , a + h2 ) ∈ , i.e.,

(τ, a) ∈ h1 ,h2 := {(a, b) ∈ R2 | a ≥ 0, b ≥ 0, a + b ≤ ω − h1 − h2 }.

Then we find that

|y(T0 , τ + h1 ; a + h2 ) − y(t, τ ; a)|

= |y(T0 − τ − h1 , 0; a + h2 ) − y(T0 − τ, 0; a)|
≤ |k1 (a + h2 ) − k1 (a)|φ(T0 − τ − h1 )x(T0 − τ − h1 , a + h2 )
+ k1 (a)|φ(T0 − τ − h1 ) − φ(T0 − τ )|x(T0 − τ − h1 , a + h2 )
+ k1 (a)φ(T0 − τ )|x(T0 − τ − h1 , a + h2 ) − x(T0 − τ, a)|
≤ |k1 (a + h2 ) − k1 (a)|C0 k2∞ ω · 1 + k1∞ ε · 1 + k1∞ · C0 k2∞ ω · C0 k2∞ ω(1 + k1∞ )ε,

where an evaluation similar to Equation (25) and relation (21) are used. Hence, we have

|y(T0 , τ + h1 ; a + h2 ) − y(t, τ ; a)| dτ da ≤ C1 ε,
h1 ,h2
462 K. Kawachi

where
ω2
C1 := C0 k2∞ ω2 + {k1∞ + C02 (k2∞ )2 k1∞ (1 + k1∞ )ω2 } .
2

It follows that

|y(T0 , τ + h1 ; a + h2 ) − y(t, τ ; a)| dτ da

 
= |y(T0 , τ + h1 ; a + h2 ) − y(t, τ ; a)| dτ da + y(t, τ ; a) dτ da
h1 ,h2 \h1 ,h2

≤ (C1 + 2)ε.

Other cases, for example, h1 < 0 and h2 > 0, are about the same, the discussion of which is
omitted here. Hence, it follows from Theorem 2.4 that P2 B is relatively compact. This completes
our proof. 

And now let us prove the result on persistence with the above preparations, which imply that
system (14) satisfies all assumptions in Theorem 2.3. Roughly speaking, the following theorem
states that the rate of infection is bounded away from 0 and the bound is eventually independent
of the initial data.

THEOREM 4.6 System (14) is uniformly strongly ρ-persistent if R0 > 1.

Proof Thanks to Theorem 2.3, we have to show only the uniform weak ρ-persistence for
system (14).
Suppose that system (14) is not uniformly weakly ρ-persistent, then for any ε ∈ (0, 1) we
can choose some initial condition (x0 , y0 ) and some sufficiently large T0 > 0 which satisfy
ρ(x0 , y0 ) > 0 and

 ω  a
k2 (a, τ )y(t, τ ; a − τ ) dτ da ≤ ε for all t ≥ T0 .
0 0

Then we obtain

λ(t, a) ≤ k1 (a)C̃(P (t))ε for all t ≥ T0 ,

which implies that

 a
x(t, a) = x(t − a, 0) exp − λ(t − a + r, r) dr
0
 a
≥ exp −ε k1 (r)C̃(P (t − a + r)) dr (26)
0

for any t ≥ T0 .
 Journal of Biological Dynamics 463

At the same time, if t > ω, then we have

 ω b
φ(t) = C̃(P (t)) k2 (b, τ )k1 (b − τ )φ(t − τ )x(t − τ, b − τ ) dτ db
0 0
 ω   ω 
= φ(t − τ ) C̃(P (t)) k2 (b, τ )k1 (b − τ )x(t − τ, b − τ ) db dτ
0 τ
 t   t 
= φ(t − τ ) C̃(P (t)) k2 (b, τ )k1 (b − τ )x(t − τ, b − τ ) db dτ,
0 τ

where we have used the fact that k2 (b, τ ) = 0 if b > ω or τ > ω, which is obvious from
Assumption 4.1 iv)b). For the same reason, φd (t) := φ(t + d) (d ≥ 0) satisfies that
 t+d   t+d 
φd (t) = φ(t + d − τ ) C̃(P (t + d)) k2 (b, τ )k1 (b − τ )x(t + d − τ, b − τ ) db dτ
0 τ
 t   t 
= φd (t − τ ) C̃(P (t + d)) k2 (b, τ )k1 (b − τ )x(t + d − τ, b − τ ) db dτ (27)
0 τ

for sufficiently large t (> ω) Equations (20), (26) and (27) imply that
 t   t  b−τ 
φd (t) ≥ φd (t − τ ) C̃(P0 ) k2 (b, τ )k1 (b − τ ) exp −ε C0 k1 (r) dr db dτ.
0 τ 0

If d is larger than a sufficiently large positive number T , we find that

 t+T   t+T  b−τ 
φd (t) ≥ φd (t − τ ) C̃(P0 ) k2 (b, τ )k1 (b − τ ) exp −ε C0 k1 (r) dr db dτ
0 τ 0
 t   τ +T  b−τ 
≥ φd (t − τ ) C̃(P0 ) k2 (b, τ )k1 (b − τ ) exp −ε C0 k1 (r) dr db dτ
0 τ 0

for all t ≥ 0. Taking Laplace transforms leads to

φ̂d (λ) ≥ φ̂d (λ)F (ε, λ, T ), (28)

where
 ∞   τ +T  b−τ 
−λτ
F (ε, λ, T ) := e C̃(P0 ) k2 (b, τ )k1 (b − τ ) exp −ε C0 k1 (r) dr db dτ.
0 τ 0

Since φd is bounded, φ̂d (λ) is defined for all λ ≥ 0. Moreover,

 ∞  ∞ 
lim F (0, 0, T ) = C̃(P0 ) k2 (b, τ )k1 (b − τ ) db dτ
T →∞ 0 τ
 ω  ω 
= C̃(P0 ) k2 (b, τ )k1 (b − τ ) db dτ
0 τ

= R0 > 1.

Since F (ε, λ, T ) is continuous, we see that F (ε, λ, T ) > 1 if ε, λ > 0 are chosen small enough
and T large enough. Then Equation (28) implies that φ̂d (λ) = 0, which means φd (t) = 0 a.e. on
[0, ∞), i.e., φ(t) = 0 for almost every t > d + T . This contradicts Lemma 4.4. 
464 K. Kawachi

5. Discussion

In this paper, we have shown a uniform strong persistence in two structured population models.
Uniform weak persistence has been proved by using a positivity result and the Laplace transform
of the force of infection φd (t), where we have seen the model’s basic reproduction ratio R0 plays
an important role. Then we have applied the theorem stating uniform weak persistence implies
uniform strong persistence under the compactness condition (C) to derive the two model’s uniform
strong persistence. In order to verify the compactness condition (C)(b) we have made use of
Fréchet–Kolmogorov’s theorem that states the compactness criteria for the subset of Lp -space.
In Section 3, we do not use φ(t), whose alternative is σ (t, (x, y, z)) = y(t) + z(t, ·) 1 .
Assumption 3.1 and γ > 0 assure that σ (t, (x, y, z)) > 0 if and only if φ0 (t) > 0. From this
point of view, the discussion in Section 3 and that in Section 4 are virtually identical.
Assuming that the upper bound of age or age-since-infection is finite facilitates the proof of
persistence, but we see no problem in that the assumption is realistic and reasonable. The case
that ω is infinite could be handled by a splitting introduced by Schappcher [2,3,10,16].

Acknowledgements
The author gratefully acknowledges the valuable, constructive remarks by two anonymous referees, which made this
paper much easier to read through.

References

[1] N. Dunford and J.T. Schwartz, Linear Operators. Part I: General Theory, John Wiley and Sons Ltd, New York, 1959.
[2] Z. Feng and H.R. Thieme, Endemic models with arbitrarily distributed periods of infection. I. Fundamental properties
of the model, SIAM J. Appl. Math. 61(3) (2000), pp. 803–833.
[3] Z. Feng, L. Rong, and R.K. Swihart, Dynamics of an age-structured metapopulation model, Natr. Resource Modeling
18(4) (2005), pp. 415–440.
[4] J.K. Hale, Theory of Functional Differential Equations, Springer, Berlin, 1977.
[5] V. Hutson and K. Schmitt, Permanence and the dynamics of biological systems, Math. Biosci. 111 (1992), pp. 1–71.
[6] H. Inaba, Endemic threshold results in an age-duration-structured population model for HIV infection, Math. Biosci.
201 (2006), pp. 15–47.
[7] K. Kawachi, Deterministic models for rumor transmission, Nonlinear Anal. Real World Appl. not yet published.
[8] M. Martcheva and G. Crispino-O’Connell, The transmission of meningococcal infection: a mathematical study,
J. Math. Anal. Appl. 283 (2003), pp. 252–275.
[9] M. Martcheva and C. Castillo-Chavez, Diseases with chronic stage in a population with varying size, Math. Biosci.
182 (2003), pp. 1–25.
[10] M. Martcheva and H.R. Thieme, Progression age enhances backward bifurcation in an epidemic model with super-
infection, J. Math. Biol. 46 (5) (2003), pp. 385–424.
[11] H.R. Thieme, Semiflows generated by Lipschitz perturbations of non-densely defined operators, Diff. Integr. Eqn.
3(6) (1990), pp. 1035–1066.
[12] ———, Uniform weak implies uniform strong persistence for non-autonomous semiflows, Proc. AMS 127 (1999),
pp. 2395–2403.
[13] ———, Uniform persistence and permanence for non-autonomous semiflows in population biology, Math. Biosci.
166 (2000), pp. 173–201.
[14] ———, Disease extinction and disease persistence in age structured epidemic models, Nonlinear Anal. 47 (2001),
pp. 6181–6194.
[15] ———, Mathematics in Population Biology, Princeton University, Princeton and Oxford, 2003.
[16] G.F. Webb, Theory of Nonlinear Age-Dependent Population Dynamics, Marcel Dekker, New York and Basel, 1985.