Mathematical Biology

J. Math. Biol. (2013) 67:185–215

DOI 10.1007/s00285-012-0546-5

Multiparametric bifurcations of an epidemiological

model with strong Allee effect

Linlin Cai · Guoting Chen · Dongmei Xiao

Received: 19 October 2011 / Revised: 16 April 2012 / Published online: 22 May 2012
© Springer-Verlag 2012

Abstract In this paper we completely study bifurcations of an epidemic model

with five parameters introduced by Hilker et al. (Am Nat 173:72–88, 2009), which
describes the joint interplay of a strong Allee effect and infectious diseases in a single
population. Existence of multiple positive equilibria and all kinds of bifurcation are
examined as well as related dynamical behavior. It is shown that the model under-
goes a series of bifurcations such as saddle-node bifurcation, pitchfork bifurcation,
Bogdanov–Takens bifurcation, degenerate Hopf bifurcation of codimension two and
degenerate elliptic type Bogdanov–Takens bifurcation of codimension three. Respec-
tive bifurcation surfaces in five-dimensional parameter spaces and related dynamical
behavior are obtained. These theoretical conclusions confirm their numerical simula-
tions and conjectures by Hilker et al., and reveal some new bifurcation phenomena
which are not observed in Hilker et al. (Am Nat 173:72–88, 2009). The rich and com-
plicated dynamics exhibit that the model is very sensitive to parameter perturbations,
which has important implications for disease control of endangered species.

Keywords Epidemic model · Allee effect · Degenerate elliptic type ·

Bogdanov–Takens bifurcation · Codimension three · Degenerate Hopf bifurcation

This research was partially supported by the National Natural Science Foundations of China
(No. 10831003 and No. 10925102), and the Program of Shanghai Subject Chief Scientists
(No. 10XD1406200). The third author also thank Université de Lille 1 for financial
support when she visited it and finished this paper.

L. Cai · D. Xiao (B)

Department of Mathematics, Shanghai Jiao Tong University, Shanghai 200240, China
e-mail: [email protected]

G. Chen
UFR de Mathématiques, UMR CNRS 8524, Université de Lille 1, 59655 Villeneuve d’Ascq, France
e-mail: [email protected]

123
186 L. Cai et al.

Mathematics Subject Classification Primary 34C23 · 34C25;

Secondary 34C37 · 92D15 · 92D30

1 Introduction

In 1931, Allee (1931) observed that many animal and plant species suffer a decrease
of population growth rate as their populations reach small sizes or low density, in par-
ticular, the population exhibits a “critical size or density”, below which the population
declines on average, and above which it increases on average. This phenomenon in
biology is called strong Allee effect, which is particularly relevant for endangered
species, small or invasive populations. It can be caused by a number of mechanisms,
for instance, higher juvenile mortality by genetic inbreeding and difficulties in find-
ing mating partners at low population densities (see Courchamp et al. 2008). On the
other hand, infectious diseases are becoming increasing recognized as a major threaten
factor in population viability (see Anderson and May 1979; Daszak et al. 1999; de
Castro and Bolker 2005; Haydon et al. 2002; Hilker et al. 2007, 2005; Hudson et al.
2001 and references therein). Hence, some species suffer from both disease and strong
Allee effect, as observed, for example, the island fox species in Clifford et al. (2006)
and Angulo et al. (2007). And the joint interplay of disease and Allee effects has been
investigated only recently by mathematical model (see Deredec and Courchamp 2006;
Friedman and Yakubu; Hilker et al. 2005, 2009; Thieme et al. 2009 and references
therein). Let us recall the mathematical model proposed by Hilker et al. for the syner-
getic interplay of Allee effects and disease infection in a host population. The reader
is referred to Hilker et al. (2009) for details about the description of the following
model,
 dx
dt = r (1 − x)(x − u)x − αy,
dy
(1.1)
dt = [−α − d − r u + (σ − 1)x − σ y]y,

where x(t) is the dimensionless total population density at time t ≥ 0 which is com-
posed of infected y(t) and susceptible x(t) − y(t), the carrying capacity is normalized
to unity, the parameter r > 0 adjusts per capita growth rate without disease, 0 < u < 1
being the Allee threshold, α > 0 being disease-induced mortality, σ > 0 being infec-
tious rate and d > 0 determines the effect of density dependence and independence
in the demographic functions.
Hilker et al. (2009) studied the existence of endemic equilibria of system (1.1)
by graphical nullcline analysis, found complicated dynamical behaviors of system
(1.1) by numerical continuation and simulations, and conjecture that these dynami-
cal behaviors are associated with Bogdanov–Takens bifurcations. Motivated by their
work, we systematically study bifurcations and dynamics of system (1.1) by theo-
retic analysis. We provide a detail qualitative and bifurcation analysis of model (1.1)
depending on all five parameters, give the necessary and sufficient condition for the
existence of endemic equilibria, clarify several threshold quantities for disease per-
sistence, extinction, and the possibility of multiple stable steady states, prove that
system (1.1) can undergo a degenerate elliptic type Bogdanov–Takens bifurcation of

123
Bifurcations of an epidemiological model 187

codimension three and a series of other bifurcations such as saddle-node bifurcation,

pitchfork bifurcation, Bogdanov–Takens bifurcation, Hopf bifurcation and degener-
ate Hopf bifurcations, and obtain the respective dynamics of system (1.1) for each
bifurcation surfaces in five-dimensional parameter spaces. These theoretical conclu-
sions confirm the conjecture of Hilker et al. (2009) and reveal some new bifurcation
phenomena, which are not observed in Hilker et al. (2009), such as pitchfork bifur-
cation, degenerate Hopf bifurcation and degenerate elliptic type Bogdanov–Takens
bifurcation of codimension three. The methods developed in this paper provide an
approach to study bifurcations and dynamics of general mathematical models with
multi-parameters.
Bifurcation theory of dynamical systems has been developed for more than half
century, which is to study the changes in the qualitative or topological structure of a
given dynamical system with parameters as the parameters vary. If a small smooth
change of parameters made to a special value of parameters (called the bifurcation
value) of a dynamical system causes a sudden ’qualitative’ or topological change in its
dynamic behavior, then we call the system undergoes a bifurcation in the small neigh-
borhood of this bifurcation value. Otherwise, we say the system is structure stable. We
refer the reader to the books Andronov et al. (1971), Chow and Hale (1982), Chow
et al. (1994), Kuznetsov (1998) for details. As an application of bifurcation theory,
bifurcation phenomenon in epidemiological models or ecological models have impor-
tant consequences for disease control or species management, respectively, (see for
example Baer et al. 2006; Etoua and Rousseau 2010; Hadeler and van den Driessche
1997; Xiao 2005; Xiao and Ruan 1999; Xiao and Zhang 2007 and references therein).
Since Bogdanov (1981) and Takens (1974) studied the 2-parameters generic family of
nilpotent cusp of codimension two and obtained a complete classification of topologi-
cal phase portraits for the family (it is usually called the Bogdanov–Takens bifurcation,
denoted by BT bifurcation for short), many researchers have devoted their effort to
generalize Bogdanov and Takens’ pioneer results to n-parameters generic family of
nilpotent equilibrium of codimension n with n ≥ 3 and there have been a series of
achievements and unprecedented challenges on the theme, for instance, degenerate
BT bifurcation of codimension 3: saddle, elliptic and focus cases (see for instance
Dumortier et al. 2001, 1987, 1991; Li and Rousseau 1989; Mardešić 1992; Xiao 1993,
2008).
In the bifurcation analysis of model (1.1), we are interested in bifurcations near the
endemic equilibria. We shall prove that the system has a nilpotent endemic equilibrium
of codimension at most 3 for all allowable parameter values, and the nilpotent endemic
equilibrium is elliptic if it exists. It is also shown that there exist three independent
parameters of (r, u, d, α, σ ) such that the system undergoes the degenerate elliptic
type BT bifurcation of codimension 3 even if it is a 5-parameters family. The bifurca-
tion diagram for this degenerate BT bifurcation of codimension 3 in Dumortier et al.
(1991) and the stability of disease-free equilibria have shown that system (1.1) has
rich bifurcation phenomenon and complicated dynamics, for instance, system (1.1)
can has homoclinic loop or two limit cycles for some values of parameters, respec-
tively. The existence of limit cycles (isolated periodic orbits) for epidemic models is
interesting and significant both in mathematics and applications since the existence of
stable limit cycle provides a satisfactory explanation for those species communities

123
188 L. Cai et al.

in which epidemic is observed to break out in a rather reproducible periodic manner.

This may have profound implications for disease control and biological conservation.
To the best of our knowledge, system (1.1) is the first example in models which
exhibits only elliptic type degenerate BT bifurcation of codimension 3 and no focus
type degenerate BT bifurcation of codimension 3. This is different from the observa-
tion in Baer et al. (2006). We shall provide explicit smooth transformations to reduce
the system with an nilpotent equilibrium into a normal form of codimension 3, which
is very useful for the bifurcation analysis. On the other hand, degenerate Hopf bifur-
cation of codimension 2 of system (1.1) is studied and the explicit conditions for the
existence of two limit cycles are given.
This paper is organized as follows. General properties of system (1.1), such as
boundedness of solutions, existence and stability of disease-free or endemic equilib-
ria, and saddle-node bifurcation are discussed in Sect. 2. In Sect. 3 we make theoretic
studies on the bifurcation of an elliptic type degenerate BT equilibrium of codimen-
sion 3, which is an endemic equilibria of system (1.1). In Sect. 4 we prove that the
system can undergo some bifurcations of codimension 2 or 1 such as BT bifurcation,
degenerate Hopf bifurcation and Hopf bifurcation, and provide an approach to study
Hopf bifurcation and degenerate Hopf bifurcation. A brief biological interpretation
and conclusion are given in the last section.

2 General analysis of system (1.1)

In this section, we make general analysis on dynamics of system (1.1). From the point
of view of biology, we only restrict our attention to system (1.1) in the closed first
quadrant in the (x, y) plane, denoted by R2+ . However, the positive y-axis (i.e. x = 0
and y > 0) is not invariant for system (1.1) and the vector fields of system (1.1) on
the positive y-axis are from right to left. Hence, the x coordinates of solution curves
crossing y-axis of system (1.1) will be negative as t increases. Based on the ecological
meaning, we adopt the following convention.

(H1): x(t) ≡ 0 for all t ≥ t0 if there exists a positive time t0 such that
the solution (x(t), y(t)) of system (1.1) satisfies x(t0 ) = 0.

We first state a lemma which shows that system (1.1) is as “well behaved” as one
intuits from biology if (H1) holds. We omit the standard proof.

Lemma 2.1 If convention (H 1) holds, then the solutions of system (1.1) are nonnega-
tive and bounded, that is, for each solution (x(t, x0 , y0 ), y(t, x0 , y0 )) of (1.1) with ini-
tial condition (x(0, x0 , y0 ), y(0, x0 , y0 )) = (x0 , y0 ) ∈ R2+ , there exists a T (x0 , y0 ) ≥
0 such that 0 ≤ x(t, x0 , y0 ) ≤ 1, 0 ≤ y(t, x0 , y0 ) ≤ max{0, σ1 (σ − 1 − α − d − r u)}
for t ≥ T (x0 , y0 ).

System (1.1) always has three disease-free (or boundary) equilibria : O(0, 0),
A(u, 0) and E 0 (1, 0) for all allowable parameters. In epidemiology, the basic repro-
duction number of an infection is a useful metric, which can be defined as the number of
secondary infections produced by a single infected during its entire infectious period in

123
Bifurcations of an epidemiological model 189

a completely susceptible population. Hilker et al. (2009) obtain the basic reproduction
number R0 and the critical host density PT for system (1.1) as follows

σ α + d + ru
R0 = , PT = .
α + d + ru + 1 σ −1

Obviously, R0 ≤ 1 if and only if σ − 1 − α − d − r u ≤ 0. And the condition

σ − 1 − α − d − r u ≤ 0 leads to the extinction of infections y(t) of system (1.1).
Hence if R0 ≤ 1, system (1.1) does not have endemic equilibria and the infectious
disease can not establish in the host population. Let us see the stability of disease-free
equilibria of system (1.1) as R0 ≤ 1. We compute the Jacobian matrix of system (1.1)
at a disease-free equilibrium (x, 0) and denote it by Jd (x, 0),
 
α f  (x) −α
Jd (x, 0) = ,
0 (σ − 1)(x − PT )

where f (x) = αr (1 − x)(x − u)x. Then the characteristic equation of Jd (x, 0) is

(λ − α f  (x))(λ − (σ − 1)(x − PT )) = 0, (2.1)

where x is the x coordinate of the disease-free equilibrium. From (2.1), we have

Theorem 2.1 If R0 ≤ 1, then system (1.1) has only three disease-free equilibria
O(0, 0), A(u, 0) and E 0 (1, 0). Both O(0, 0) and E 0 (1, 0) are stable nodes, and
A(u, 0) is a saddle. Bistable appears for system (1.1).

From Theorem 2.1 we can see whether or not the disease-free host population
goes to extinction depending on the initial population number. If the initial population
number is in the left of the stable manifold of A(u, 0) then it will go to extinction,
otherwise, it will be survival. Thus, if R0 ≤ 1 the infectious disease could not affect
the dynamics of system (1.1) which is determined only by Allee effect. This leads to
bistable phenomenon.
Note that R0 ≤ 1 is equivalent to PT ≥ 1 if σ > 1. By the formula of R0 (or PT ), we
know that R0 (PT , respectively) is a decreasing (an increasing, respectively) function
of the Allee threshold u. Thus, the basic reproduction number R0 could lessen to one
or less than one if let u increase starting at zero and other parameters keep unchanged.
Thus, R0 for the host with Allee effect is less than that for the host without Allee
effect. Theorem 2.1 shows that the disease can not establish in the host population if
R0 ≤ 1. Biologically, this conclusion shows that the presence of an Allee effect in
host populations might play a stabilizing and protective role in the invasion of disease.
And the critical host density PT ≥ 1 which prevents the population from reaching the
necessary density for disease establishment even though the infectious rate σ is not
small.
In the rest of our study, we shall focus on the ecologically more interesting case
that R0 > 1 (i.e. σ > α + d + r u + 1) and the Allee threshold is far from the car-
rying capacity, that is, 0 < u < 21 . However, R0 > 1 is equivalent to 0 < PT < 1.
Henceforth, we always assume that

123
190 L. Cai et al.

(a) (b) (c)

Fig. 1 The tangent points of the nontrivial nullclines C and L

1
(H2): 0<u< , 0 < PT < 1.
2
Under hypothetical condition (H2), we consider the existence of endemic equilibria
(i.e. positive equilibria) for system (1.1). Even though it is algebraically impossible to
find explicit sufficient and necessary conditions for the number of endemic equilibria
of system (1.1) depending on all parameters, Hilker et al. (2009) obtained the exis-
tence and number of endemic equilibria by graphical nullcline. Here we further make
theoretic analysis of system (1.1) and give a complete classification on the number of
endemic equilibria and the corresponding explicit sufficient and necessary conditions.
This is essential for study bifurcations of system (1.1).
Note that the existence of endemic equilibria of system (1.1) is equivalent to that of
intersection points of the nontrivial nullclines C and L of system (1.1) in the interior
of the first quadrant R2+ , where
 r 
C = (x, y) : y = f (x), f (x) = (1 − x)(x − u)x, 0 ≤ x < +∞ ,
 α 
σ −1
L = (x, y) : y = g(x), g(x) = (x − PT ), 0 ≤ x < +∞ .
σ

We first make analysis on the geometric character of the cubic curve C. The cubic
curve C crosses the x-axis at three points O(0, 0), A(u, 0) and E 0 (1, 0), which are
disease-free equilibria of system (1.1). And the curve C has a maximum point, a min-
imum point and a unique inflection at E ∗ (x ∗ , y ∗ ), where x ∗ = u+1 ∗
3 , y = 27α (u +
r

1)(1 − 2u)(2 − u).

Hence, E(x, y) is an endemic equilibrium of system (1.1) if and only if E(x, y)
is an intersection point of the cubic curve C and the straight line L in the interval
u < x < 1. And E ∗ (x ∗ , y ∗ ) is a unique endemic equilibrium of system (1.1) if the
straight line L is a tangent line of C passing through the inflection point E ∗ , which
intersects x-axis at (Tu , 0) and E ∗ is a triple tangent point of L and C, see Fig. 1a,
where Tu = 9(u(u+1)
3
2 −u+1) . Note that 0 < u < 2 . Thus u < Tu < 1.
1

Since E ∗ divides the curve C in the interval u ≤ x < 1 into two parts, we let γ1 be
the portion of the curve C with u ≤ x < x ∗ , and let γ2 be the portion of the curve C
with x ∗ < x < 1.

123
Bifurcations of an epidemiological model 191

We further consider if the curve C and the line L has a double point with u < x < 1.
It is clear that there are no double points of L and C if the line L crosses x-axis in
the interval Tu < x < 1. Thus, L and C has a unique cross point in the interval
u < x < 1 if Tu < PT < 1. And the line L may be tangent to the curve C in the
interval u < x < 1 only if 0 < PT < Tu . Next we seek the conditions of being tangent
between the line L and the curve C in the interval u < x < 1.
Suppose that the line L is tangent to the curve C in the interval u < x < 1. Then
the following equations


f (x) = g(x) > 0,
(2.2)
f  (x) = σ σ−1

must have solutions in the interval u < x < 1.

It is clear that the existence of solutions to (2.2) in the interval u < x < 1 is
equivalent to that of the following equation

f (x) − f  (x)(x − PT ) = 0, 0 < PT < Tu , (2.3)

where f  (x) = − αr 3x 2 − 2(1 + u)x + u .

By Cardan formula, we compute the discriminant  of Eq. (2.3) and obtain that

PT
= (1 − PT )(u − PT )(Tu − PT ).
3888(u 2 − u + 1)

Hence, there exists a unique real root x0 with u < x0 < 1 of Eq. (2.3) if 0 < PT < u.
Since  = 0 as PT = u, Eq. (2.3) has two different real solutions u and 21 . And if
u < PT < Tu , then  < 0 and Eq. (2.3) has only two different roots x01 and x02 with
u < x02 < u+1 3 < x 01 < 1 by computation.
When 0 < PT < u, by the geometric character of C and L, we obtain that in
the interval u < x < 1 the line L and the curve C has only a point of tangency at
E x0 = (x0 , f (x0 )) ∈ γ2 if σ σ−1 = f  (x0 ).
When PT = u, there are two cases for the points of tangency for L and C in
u < x < 1. One case is that the line L and the curve C has only a point of tangency
at E A1 = ( 21 , 8α
r
(1 − 2u)) ∈ γ2 if σ σ−1 = 4α r
, see Fig. 1b. And the other case is that
the line L and the curve C has a point of tangency at A(u, 0) and a cross point at
E A2 = (1 − u, αr u(1 − 2u)(1 − u)) ∈ γ2 if σ σ−1 = r u(1−u) α .
When u < PT < Tu , by calculation we find that there are also two cases for the
points of tangency for L and C in u < x < 1. If σ σ−1 = f  (x01 ), then the line L
and the curve C has a point of tangency at E x011 = (x01 , f (x01 )) ∈ γ2 and a cross
point at E x012 = (x012 , f (x012 )) ∈ γ1 , where x012 is a simple root of the equation
f (x) − f  (x01 )(x − x0 ) = 0.
If σ σ−1 = f  (x02 ), then the line L and the curve C has a point of tangency at
E x021 = (x02 , f (x02 )) ∈ γ1 and a cross point at E x022 = (x022 , f (x022 )) ∈ γ2 , see
Fig. 1c, where x022 is a simple root of the equation f (x) − f  (x02 )(x − x0 ) = 0.

123
192 L. Cai et al.

Note that

r u(1 − u) r 2 r r 2
< f  (x02 ) < (u − u + 1), < f  (x01 ) < (u − u + 1),
α 3α 4α 3α

and f  (x02 ) < f  (x01 ). Therefore, the line L and the curve C have three different
cross points in the interval u < x < 1 if f  (x02 ) < σ σ−1 < f  (x01 ).
Summing up the above analysis, we have the following lemma on the existence and
non-existence of endemic equilibria of system (1.1).

Lemma 2.2 Assume that (H 2) holds. Then system (1.1) has at most three endemic
equilibria and at least no endemic equilibria. More precisely,

(i) system (1.1) has three endemic equilibria if and only if

σ −1
f  (x02 ) < < f  (x01 ), u < PT < Tu . (2.4)
σ

The three endemic equilibria are cross points of the line L and the curve C.
(ii) System (1.1) has two endemic equilibria if and only if one of the following
conditions holds.

σ −1
0< < f  (x0 ), 0 < PT < u; (2.5)
σ
r u(1 − u) σ −1 r
< < , PT = u; (2.6)
α σ 4α
σ −1
= f  (x01 ), u < PT < Tu ; (2.7)
σ
σ −1
= f  (x02 ), u < PT < Tu . (2.8)
σ

The two endemic equilibria are cross points of L and C if one of (2.5) and (2.6)
holds. And if one of (2.7) and (2.8) holds, then one endemic equilibrium is a
point of tangency of L and C and the other is a cross point of L and C.
(iii) System (1.1) has one endemic equilibrium if and only if one of the following
conditions holds.

Tu < PT < 1; (2.9)

σ −1   σ −1
either < f (x02 ) or f (x01 ) < , and u < PT < Tu ;
σ σ
(2.10)
σ −1 r u(1 − u)
= , PT = u; (2.11)
σ α
σ −1 r 2
= (u − u + 1), PT = Tu ; (2.12)
σ 3α

123
Bifurcations of an epidemiological model 193

σ −1
= f  (x0 ), 0 < PT < u; (2.13)
σ
σ −1 r
= , PT = u; (2.14)
σ 4α
If one of (2.9)–(2.11) holds, then the unique endemic equilibrium is a cross point
of L and C. If one of (2.12)–(2.14) holds, then the unique endemic equilibrium
is a tangency point of L and C, which is at E ∗ , E x0 and E A1 , respectively.
(iv) System (1.1) has no endemic equilibria if and only if

σ −1
> f  (x0 ), 0 < PT ≤ u, (2.15)
σ
where f  (x0 ) = r
4α as PT = u.
In what follows we study the stability of equilibria and phase portraits of sys-
tem (1.1). First we consider the stability of disease-free equilibria for system (1.1) if
0 < PT < 1. From the eigenvalues of (2.1), we have the following.
Lemma 2.3 Disease-free equilibria of system (1.1) has the following stabilities,
respectively.
(i) O(0, 0) is a stable hyperbolic node and E 0 (1, 0) is a hyperbolic saddle if 0 <
PT < 1;
(ii) A(u, 0) is a unstable hyperbolic node if 0 < PT < u; A(u, 0) is a hyperbolic
saddle if u < PT < 1 and A(u, 0) is a saddle-node if PT = u.
In Table 2 of Hilker et al. (2009), Hilker et al. claimed that E 0 (1, 0) is a unstable
node if 0 < PT < 1. This is not true. From Lemma 2.3, we know that E 0 (1, 0) is a
hyperbolic saddle if 0 < PT < 1. Further we obtain the following global conclusions
on dynamics of system (1.1).
Theorem 2.2 If (H 2) and (2.15) hold, then system (1.1) has only disease-free equi-
libria and no endemic equilibria. Disease-free equilibria O(0, 0) is a stable node,
A(u, 0) is a unstable node or saddle-node, and E 0 (1, 0) is a hyperbolic saddle. Almost
all orbits of system (1.1) in the interior of R2+ tend to O(0, 0) (see Fig. 2).
Biologically, this conclusion shows that the joint interplay between infectious dis-
eases and Allee effects is deathblow for the host population if the critical host density
PT is low and the infectious rate σ is large. This leads that the whole population goes
to extinction.
In the following we investigate the stability of endemic equilibria of system (1.1).
We denote the Jacobian matrix  of system (1.1)at an endemic equilibrium E(x, y) by
α f  (x) −α
Je (x, y), where Je (x, y) = . Then the characteristic equation of
(σ − 1)y −σ y
Je (x, y) is
 
σ −1
λ2 + (σ f (x) − α f  (x))λ − ασ f (x) f  (x) − = 0, (2.16)
σ

where x is the x coordinate of the endemic equilibrium.

123
194 L. Cai et al.

Fig. 2 Extinction of the population in system (1.1) as R0 > 1, PT = u and σ σ−1 > 4α
r , where the values
of parameters for the figure are r = 1, u = 0.3, d = 0.2, α = 1, σ = 6

From the distribution of eigenvalues of characteristic equation (2.16), we now derive

the topological nature of endemic equilibria of system (1.1) as follows.

Theorem 2.3 The endemic equilibrium E(x, y) is degenerate if it is a tangent point

of L and C. Otherwise, the endemic equilibrium E(x, y) is an elementary equilib-
rium. More precisely, the endemic equilibrium E(x, y) is a center or weak focus if
f  (x) − σ σ−1 < 0 and σ f (x) − α f  (x) = 0; the endemic equilibrium is a hyper-
bolic node or focus if f  (x) − σ σ−1 < 0 and σ f (x) − α f  (x) = 0, and the endemic
equilibrium is a hyperbolic saddle if f  (x) − σ σ−1 > 0.

From Lemma 2.2 and Theorem 2.3, we easily check that the parameters satisfy-
ing condition (2.7), (2.8), (2.13) or (2.14), respectively come to being a hypersurface
in five-dimensional parameter space (α, d, r, u, σ ). These hypersurfaces are saddle-
node bifurcation surfaces of system (1.1). As parameters (α, d, r, u, σ ) vary in the
small neighborhood of the these hypersurfaces and cross the respective hypersurfaces,
system (1.1) undergoes saddle-node bifurcation. Moreover, the parameters satisfying
condition (2.12) consist of a hypersurface, which is pitchfork bifurcation surface. As
parameters (α, d, r, u, σ ) vary in the small neighborhood of this hypersurface and
transversal to it, system (1.1) undergoes pitchfork bifurcation. Let us see the Fig. 3.
We choose PT and σ σ−1 as two bifurcation parameters. When (PT , σ σ−1 ) vary in the
range
  
σ −1
1 = PT , : Tu ≤ PT < 1, σ > ,
σ

system (1.1) always has an endemic equilibrium. And as bifurcation parameters

(PT , σ σ−1 ) pass through the point Q = (Tu , 3α
r
(u 2 − u + 1)) and go into the domain
2 bounded by L a , L b and PT = u, system (1.1) has three endemic equilibria.

123
Bifurcations of an epidemiological model 195

Fig. 3 The bifurcation diagram

of equilibria of system (1.1),
where PT and σ σ−1 are
bifurcation parameters,
r (u 2 − u + 1)) is a
Q(Tu , 3α
bifurcation point, 1 EE
represents “one endemic
equilibrium” and 3 EE means
“three endemic equilibria”

This implies that endemic equilibria of system (1.1) always exist as (α, d, r, u, σ )
vary in the direction from 1 through Q into 2 , or vice versa. Hence, in the case the
disease can establish in the host population.

3 Elliptic type degenerate BT bifurcation of codimension three

In this section we first give a lemma, which provides a series of explicit smooth trans-
formations to derive a normal form with terms up to and including fourth order for
the codimension three BT equilibrium. Then we apply the lemma to system (1.1) and
obtain that the system has a nilpotent equilibrium of codimension at most 3 for all
allowable parameters. Further we verify that the degenerate equilibrium is an elliptic
type equilibrium of codimension 3 if it exists. Last we choose three independent param-
eters of (r, u, d, α, σ ) such that the system undergoes the degenerate BT bifurcation
of codimension 3 in the elliptic case.
Note that Xiao and Ruan (1999) have provided a series of explicit smooth trans-
formations to derive a normal form with terms up to and including third order for a
nilpotent (or double-zero eigenvalue) equilibrium of planar systems and obtain the
normal form of nilpotent equilibrium of codimension 2. Here we follow their method
to provide a series of explicit smooth transformations to derive a normal form for the
nilpotent equilibrium of codimension 3.
Consider the following system
⎧
⎪
⎪ ẋ = y + a30 x 3 + a03 y 3 + a21 x 2 y + a12 x y 2 + a40 x 4
⎨
+a04 y 4 + a31 x 3 y + a13 x y 3 + a22 x 2 y 2 + R1 (x, y),
(3.1)
⎪
⎪ ẏ = b11 x y + b30 x 3 + b03 y 3 + b21 x 2 y + b12 x y 2 + b40 x 4
⎩
+b04 y 4 + b31 x 3 y + b13 x y 3 + b22 x 2 y 2 + R2 (x, y),

where b11 = 0, ai j , bi j , i, j = 0, . . . , 4, 3 ≤ i + j ≤ 4 are real parameters, and

R1 (x, y) and R2 (x, y) are smooth functions of their arguments with at least fifth-order
terms of (x, y). Then we have

123
196 L. Cai et al.

Lemma 3.1 Assume that b11 b30 = 0. Then there exists a small neighborhood U0 of
(0, 0) such that in this neighborhood U0 system (3.1) is locally topologically equivalent
to
⎧
⎨ ẋ = y,
ẏ = b11 x y + b30 x 3 + (b21 + 3a30 )x 2 y + (b40 − b11 a30 )x 4 (3.2)
⎩
+(4 a40 + b31 + 13 b11 a21 + 16 b11 b12 )x 3 y + R(x, y),

where R(x, y) is a smooth function of their arguments with at least fifth-order terms
of (x, y).
Moreover, assume that 5b30 (b21 + 3a30 ) − 3b11 (b40 − b11 a30 ) = 0. Then the equi-
librium (0, 0) of system (3.1) is a degenerate saddle of codimension 3 if b30 > 0; it is
a degenerate focus or center of codimension 3 if b30 < 0 and b11 2 + 8b < 0; and it
30
is a degenerate elliptic of codimension 3 if b30 < 0 and b11 + 8b30 ≥ 0.
2

Proof It is clear that (0, 0) is an equilibrium of system (3.1) with two zero eigenvalues.
We perform a near-identity smooth change of coordinates

X = x − ( 13 a21 + 16 b12 )x 3 − ( 21 a12 + 21 b03 )x 2 y,
Y = y + a30 x 3 + a03 y 3 − 21 b12 x 2 y − b03 x y 2 .

This transformation is invertible in a small neighborhood of the origin, and its inverse
can be found by the method of unknown coefficients. Thus, system (3.1) becomes
⎧
⎪
⎪ Ẋ = Y + a40 X 4 + a04 Y 4 + [a31 − 21 b11 (a12 + b03 )]X 3 Y
⎪
⎪
⎨ +a13 X Y 3 + a22 X 2 Y 2 + R3 (X, Y ),
Ẏ = b11 X Y + b30 X 3 + (b21 + 3a30 )X 2 Y + (b40 − b11 a30 )X 4 (3.3)
⎪
⎪
⎪
⎪ +[b31 − b11 (− 13 a21 − 16 b12 )]X 3 Y + (b13 + 2b11 a03 )X Y 3
⎩
+b04 Y 4 + [b22 − 21 b11 b03 + 21 b11 a12 ]X 2 Y 2 + R4 (X, Y ),

where R3 (X, Y ) and R4 (X, Y ) are smooth functions of their arguments with at least
fifth-order terms of (X, Y ).
In order to kill non-resonant quartic terms of system (3.3), we let
⎧
⎪
⎪ z 1 = X + − 41 a31 − 12 1
b11 a12 + 16 b11 b03 − 12
1
b22 X 4 + a04 Y 4
⎨
− 3 a22 + 6 b13 X Y − ( 2 a13 + 2 b04 )X Y 2 ,
1 1 3 1 1 2

⎪
⎪ z = Y + a40 X 4 + (− 13 b22 + 16 b11 b03 − 16 b11 a12 )X 3 Y
⎩ 2
−b04 X Y 3 − ( 21 b13 + b11 a03 )X 2 Y 2

in the small neighborhood of the origin. System (3.3) is transformed into

⎧
⎨ z˙1 = z 2 + R5 (z 1 , z 2 ),
z˙2 = b11 z 1 z 2 + b30 z 13 + (b21 + 3a30 )z 12 z 2 + (b40 − b11 a30 )z 14 (3.4)
⎩
+(4 a40 + b31 + 13 b11 a21 + 16 b11 b12 )z 13 z 2 + R6 (z 1 , z 2 ),

where R5 (z 1 , z 2 ) and R6 (z 1 , z 2 ) are smooth functions of their arguments with at least

fifth-order terms of (z 1 , z 2 ).

123
Bifurcations of an epidemiological model 197

Last we set x = z 1 , y = z 2 + R5 (z 1 , z 2 ), system (3.4) becomes

⎧
⎨ ẋ = y,
ẏ = b11 x y + b30 x 3 + (b21 + 3a30 )x 2 y + (b40 − b11 a30 )x 4 (3.5)
⎩
+(4 a40 + b31 + 13 b11 a21 + 16 b11 b12 )x 3 y + R(x, y),

which implies that system (3.1) is locally topologically equivalent to system (3.2) in
a neighborhood of the origin.
From arguments in Dumortier et al. (1991), we know that the equilibrium (0, 0)
of system (3.2) is a degenerate equilibrium of codimension 3 if 5b30 (b21 + 3a30 ) −
3b11 (b40 − b11 a30 ) = 0. And according to Theorem 7.2 in Zhang et al. 1992, p. 152,
we obtain the topological classification of (0, 0) of system (3.1) as follows: the equilib-
rium (0, 0) is a degenerate saddle if b30 > 0; it is a degenerate focus or center if b30 < 0
2 + 8b
30 < 0; and it is a degenerate elliptic if b30 < 0 and b11 + 8b30 ≥ 0.
and b11 2

Therefore, the proof is completed.

Now we discuss the topological classification of the degenerate equilibrium E ∗ of

system (1.1).

Theorem 3.1 Assume that PT = Tu and σ σ−1 = 3α r

(u 2 − u + 1). Then system (1.1)
∗ ∗ ∗
has a unique endemic equilibrium E = (x , y ), where

u+1 r
x∗ = , y∗ = (u + 1)(1 − 2u)(2 − u).
3 27α

The endemic equilibrium E ∗ is a unstable (stable) degenerate node of codimension

one if σα − u+1 α ∗
3 + PT > 0 ( σ − 3 + PT < 0, respectively), and E is an elliptic
u+1
α
equilibrium of codimension 3 if σ − 3 + PT = 0.
u+1

Proof From Lemma 2.2 we know that system (1.1) has a unique endemic equilibrium
E ∗ = (x ∗ , y ∗ ) if PT = Tu and σ σ−1 = 3α
r
(u 2 − u + 1).
Moving E to the origin, we set x1 = x − x ∗ , y1 = y − y ∗ . Note that PT = Tu
∗

and σ σ−1 = 3α r
(u 2 − u + 1). Then system (1.1) becomes

 α(σ −1)
ẋ1 = σ x1 − αy
 1 − r2 x1 ,
3

(σ −1) (3.6)
ẏ1 = ( 3 − PT )
u+1
σ x1 − (σ − 1)y1 + (σ − 1)x1 y1 − σ y12 ,

We see that the Jacobian matrix at the equilibrium (0, 0) of system (3.6) has two eigen-
values 1 and 2 , 1 = 0 and 2 = σα − u+1 3 + PT . Therefore, we distinguish two
cases 2 = 0 and 2 = 0 to discuss the topological classification of the equilibrium
(0, 0).
If 2 = 0, i.e. σα − u+1
3 + PT = 0, then by standard center manifold theory we can
obtain that endemic equilibrium E ∗ of system (1.1) is a unstable (stable) degenerate
node if σα − u+1 α
3 + PT > 0 ( σ − 3 + PT < 0, respectively).
u+1

123
198 L. Cai et al.

α
If 2 = 0, i.e. σ − u+1
3 + PT = 0, then system (3.6) can be written as

α(σ −1)
x˙1 = σ x1 − αy1 − r x13 ,
α(σ −1) (3.7)
x1 − α(σσ−1) y1 + (σ
2
y˙1 = σ2
− 1)x1 y1 − σ y12 .

Equilibrium (0, 0) of system (3.7) has double zero eigenvalues. In the following we
reduce system (3.7) to a normal form of Bogdanov–Takens singularity. Let

α(σ − 1)
x2 = x1 , y2 = x1 − αy1 .
σ

Then system (3.7) becomes

σ 2 αr (σ −1) 3
ẋ2 = y2 − r x23 , ẏ2 = α y2 − (σ − 1)x2 y2 − σ x2 . (3.8)

In a small neighborhood of the origin, we make a near-identity topological trans-

formation of coordinates x3 = x2 − 21 σα x22 , y3 = y2 − σα x2 y2 , which transforms
system (3.8) into
⎧ 3 σ2 2 rσ 4 σ3 3
⎨ ẋ3 = y3 − 2 α 2 x3 y3 − r x3 − 2α x3 − 2α 3 x3 y3 + O((x3 , y3 ) ),
⎪ 3 5

ẏ3 = −(σ − 1)x3 y3 − σα 2 x3 y32 − σ (σ2α−1) x32 y3 − 21 r (σ − 1)x34

2
(3.9)
⎪
⎩
− αr (σσ−1) x33 + σ (2αr +3σ
2 −3σ )
x33 y3 − 25 σα 3 x32 y32 + O((x3 , y3 )5 ).
3
2α 2

From Lemma 3.1, we can reduce system (3.9) to the following system
⎧
⎪
⎪ ẋ = 
y,
⎨ 
ẏ = σ (σαr−1) x y − x 3 + [ α(σ3σ−1) − 2ασ 2 r ]x 2 y + 23 σα αr (σσ−1) x 4
2
(3.10)
⎪
⎪ 
⎩ +[ 23 ασ3 r − α 2 2σ
3 2
] αr (σσ−1) x 3 y + O(|x, y|5 ).
(σ −1)

 
Since −5( α(σ3σ−1) − σ2
2r α 2
) − 3 σ (σαr−1) 3σ σ σ 15
2α αr (σ −1) = − α ( σ −1 + αr )
2σ
= 0, the
equilibrium (0, 0) of system (3.10) is a degenerate equilbrium of codimension 3 by
Lemma 3.1.
Furthermore, we claim that equilibrium (0, 0) of system (3.10) is an elliptic degen-
erate equilibrium of codimension 3.
Indeed, we only need to prove that σ (σαr−1) − 8 ≥ 0 by Lemma 3.1. From the con-
ditions PT = Tu , σ σ−1 = 3α
r
(u 2 − u + 1) and σα − u+1
3 + PT = 0, we can solve the
three equations and obtain that

(σ −1)(2−u)(1−2u)(u+1)
r = 3(u 2 −u+1)2
, α = σ (2−u)(1−2u)(u+1)
9(u 2 −u+1)
,
d= − 9(u(u+1)
2 −u+1)2 σ (1 + u)2 (u 2 − 4u + 1) + u 4 − 5u 3 + 15u 2 − 5u + 1 .
(3.11)

123
Bifurcations of an epidemiological model 199

Obvious, r and α are positive if σ > 1 and 0 < u < 21 , and d is positive for all
allowance u and σ by biological meaning.
Assume to the contrary that σ (σαr−1) < 8. Then (u+1)
27(u 2 −u+1)3
2 (u−2)2 (2u−1)2 − 8 < 0, which
√
leads to 0 < u < 27 − 3 2 5 . We now prove that this implies that d < 0.
In fact, from the expression of d we know that h(u)d < 0, where

h(u) = σ (1 + u)2 (u 2 − 4u + 1) + u 4 − 5u 3 + 15u 2 − 5u + 1.

√
Note that σ > 1. If 0 < u < 7
2 − 3 5
2 , we then have

√ √
h(u) = (σ − 1)(u + 1)2 (2 + 3 − u)(2 − 3 − u) + 2u 4 − 7u 3 + 9u 2 − 7u + 2

> 2u 4 − 7u 3 + 9u 2 − 7u + 2 = q(u).

√
Since q  (u) = 24u 2 − 42u + 18 = 6(3 − 4u)(1 − u) > 0, q  (u) < q  ( 27 − 3 2 5 ) < 0.
Hence, q(u) > q(0) > 0. This leads to that h(u) > 0 and consequently, d < 0. √
However, we know that d must be positive for biological meaning. So u ≥ 27 − 3 2 5 ,
which implies σ (σαr−1) ≥ 8. We complete the proof.

Since PT = Tu , PT > u. From Lemma 2.3 and Theorem 3.1, we obtain the dynam-
ics of system (1.1) with a triple endemic equilibrium as follows.

Theorem 3.2 Assume that PT = Tu and σ σ−1 = 3α r

(u 2 − u + 1). Then u < PT < 1
and system (1.1) has a unique endemic equilibrium E ∗ = (x ∗ , y ∗ ) and three disease-
free equilibria: O(0, 0), A(u, 0) and E 0 (1, 0). O(0, 0) is a stable node, A(u, 0) and
E 0 (1, 0) are hyperbolic saddles. The endemic equilibrium E ∗ is a degenerate unstable
(stable) node if σα − u+1 α
3 + PT > 0 ( σ − 3 + PT < 0, respectively), and E is an
u+1 ∗

elliptic equilibrium of codimension 3 if (r, d, α, u, σ ) satisfies (3.11). The dynamics

of system (1.1) in this case is shown in Fig. 4.
As observed in Hadeler and van den Driessche (1997) and Hilker et al. (2009), the
basic reproduction number is not the unique indicator for disease persistent. Theorem
3.2 shows if the disease can persist depends on the initial infectious number, Allee
threshold and the infectious rate even though the basic reproduction number R0 > 1
(i.e. 0 < PT < 1). Disease establishment and bi-stability occur if u < PT < 1 and
α
σ ≤ 3 − PT , and the whole population can go extinction if u < PT < 1 and
u+1
α
σ > 3 − PT .
u+1

In the following we study if system (1.1) can undergo degenerate BT bifurcation

of codimension 3 in a small neighborhood of equilibrium E ∗ (x ∗ , y ∗ ) as parameters
(r, d, α, u, σ ) varies in a small neighborhood of (r0 , d0 , α0 , u 0 , σ0 ) which satisfies
(3.11).
We choose (d, u, σ ) as bifurcation parameters. Let

d = d0 + λ1 , u = u 0 + λ2 , σ = σ0 + λ3

123
200 L. Cai et al.

Fig. 4 The dynamics of system (1.1) with an elliptic type endemic equilibrium of codimension 3, where
r = 2, u = 0.45, d = 3.05, α = 0.55 and σ = 11. The endemic equilibrium E ∗ is ( 21 , 22 1 ), which is
represented by ∗

where λ = (λ1 , λ2 , λ3 ) is a very small parameter vector (0 < ||λ|| 1), and
(d0 , u 0 , σ0 , α0 , r0 ) is a set of given parameters which satisfy (3.11). For simplification
we denote (d0 , u 0 , σ0 , α0 , r0 ) by (d, u, σ, α, r ).
Now we rewrite system (1.1) as follows.

ẋ = r (1 − x)(x − u − λ2 )x − αy,
(3.12)
ẏ = [−α − d − λ1 − r (u + λ2 ) + (σ + λ3 − 1)x − (σ + λ3 )y]y.

Let x1 = x − x ∗ , y1 = y − y ∗ . Then by (3.11) system (3.12) becomes


ẋ1 = c1 + ax1 + by1 + p11 x12 − r x13 ,
(3.13)
ẏ1 = c2 + cx1 + ey1 + 2q12 x1 y1 + q22 y12 ,

where
α(σ −1)
a= σ + r λ32 (2u − 1), b =−α, c = (σ + λ3 − 1) α(σσ−1)
2 , 
α(σ −1)
c1 = 19 r λ2 (u + 1)(u − 2), c2 = −λ1 − r λ2 + 3 − σ 2
u+1
λ3 α(σσ−1)
2 ,
   
α(σ −1) −1)
e=− σ + λ1 + r λ2 − u+1
3 − 2α(σ
σ2
λ3 ,
σ +λ3 −1
p11 = r λ2 , q12 = 2 , q22 = −(σ + λ3 ).

By an affine transformation x2 = x1 , y2 = c1 + ax1 + by1 , system (3.13) becomes

⎧
⎪ ẋ2 = y2 + p11 x22 − r x23 , 
⎪
⎪
⎨ ẏ = bc − ec + q22 c1 2 + bc − ae − 2q c + 2 q22 c1 a x
2 2 1 b 12 1 b 2
(3.14)
⎪
⎪ q c
+(a + e − 2 22b 1 )y2 + (ap11 − 2aq12 + 22b )x22q a 2
⎪
⎩
+ qb22 y22 + 2 q12 − q22b a x2 y2 − ar x23 .

123
Bifurcations of an epidemiological model 201

In a small neighborhood of the origin, we let

r r
x3 = x2 + x 2, y3 = y2 + p11 x22 + x2 y2 .
2 p11 2 p11

Then system (3.14) can be converted into

⎧
⎨ ẋ3 = y3 + Q 1 (x3 , y3 , λ),
ẏ = c0 + c10 x3 + c20 x32 + c30 x33 + y3 (c01 + c11 x3 + c21 x32 + c31 x33 )
⎩ 3
+Q 2 (x3 , y3 , λ),
(3.15)

where Q i (x3 , y3 , λ) = O((|x3 , y3 | + |λ|)4 ), i = 1, 2, and c0 and ci j are smooth func-

tions of (r, u, α, d, σ, λ). We omit their long expressions here for reasons of space.
Following the steps in Dumortier et al. (1991), system (3.15) can be further trans-
formed to the following canonical unfolding.
⎧
⎨ ẋ = y,
ẏ = −x 3 + μ2 (λ)x + μ1 (λ) + y(μ3 (λ) + b(λ)x + c(λ)x 2 (3.16)
⎩
+x 3 h(x, λ)) + y 2 Q(x, y, λ),

where μi (λ), b(λ),

 c(λ), h(x, λ) and Q(x, y, λ) are smooth functions of their argu-
ments, b(0) = σ (σαr−1) , c(0) = α(σ3σ−1) − 2ασ 2 r , and the determinant of the Jacobian
2

matrix of (μ1 (λ), μ2 (λ), μ3 (λ)) with respect to (λ1 , λ2 , λ3 ) at (0, 0, 0) is as follows,
 
 1 ,μ2 ,μ3 )  α 3 r (σ −1)2 (u+1)(2−u) (1−2u)2 (σ −1)+(u+1)(2−u)
D =  ∂(μ
∂(λ1 ,λ2 ,λ3 )  = 27σ 4 (u 2 −u+1)
= 0.
(0,0,0)

Hence, system (3.12) can undergo degenerate BT bifurcation of codimension 3

in the elliptic case by conclusions in Dumortier et al. (1991). Moreover, by blowing
up techniques and the integral factor, system (3.15) can be considered as a perturba-
tion of Hamiltonian systems, in which there exist a Hamiltonian system such that the
level sets of the Hamiltonian function are ovals surrounding a center. The ovals can
be bifurcated at most two limit cycles by property of Abelian integral, and the limit
cycles surround only one equilibrium (see Zoladek’s work in Dumortier et al. 1991).
Therefore, we have the following.

Theorem 3.3 System (1.1) undergoes degenerate BT bifurcation (elliptic case) of co-
dimension 3 in a small neighborhood of the equilibrium E ∗ (x ∗ , y ∗ ) if we fix parameters
u = u 0 , σ = σ0 , d = d0 and let parameters (d, u, σ ) vary in a small neighborhood
of (d0 , u 0 , σ0 ), where (d0 , u 0 , σ0 , r0 , α0 ) satisfies (3.11). More precisely, there exist
parameter values (d, u, σ ) such that system (1.1) has three endemic equilibria: one is
hyperbolic saddle and the other two are focus or nodes, and a homoclinic loop; there
exist other parameter values (d, u, σ ) such that system (1.1) has only one endemic
equilibrium and at most two limit cycles.

123
202 L. Cai et al.

Fig. 5 The dynamics of system (1.1) with three endemic equilibria, where r = 5, u = 0.15, d = 0.0813,
α = 1.441 and σ = 13.5899. The three endemic equilibria are approximately (0.257, 0.071),
(0.337, 0.145) and (0.555, 0.347), which are represented by ∗

From Theorem 3.3, we know that parameters satisfying condition (3.11) form a
surface DBT in five-dimensional parameter space (α, d, r, u, σ ), where
⎧ ⎫
⎪
⎪ α = σ (2−u)(1−2u)(u+1)
2 −u+1) , ⎪
⎪
⎪
⎪
9(u ⎪
⎪
⎪
⎪ (u+1) ⎪
⎪
⎪
⎪ d = − σ (1 + u)2 (u 2 − 4u + 1) ⎪
⎪
⎨ 9(u −u+1)
2 2
⎬
DBT = (α, d, r, u, σ ) : +u − 5u + 15u − 5u + 1 > 0,
4 3 2 ,
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ r = (σ −1)(2−u)(1−2u)(u+1) , ⎪
⎪
⎪
⎪ 3(u 2 −u+1)2 ⎪
⎪
⎩ ⎭
σ > 1, 2 > u > 0
1

which is a degenerate BT bifurcation surface of codimension 3 in the elliptic case. As

parameters (α, d, r, u, σ ) are in DBT , dynamics of system (1.1) is shown in Fig. 4.
If parameters (α, d, r, u, σ ) vary in a small neighborhood of DBT , then system (1.1)
has different dynamics depending on parameter values as described in Theorem 3.3.
In the case, dynamics of (1.1) depends on not only parameters but also the initial con-
ditions. Biologically, this conclusion shows that the host population with the strong
Allee effect is very fragile as parameters (α, d, r, u, σ ) vary in a small neighborhood of
DBT . There are many outcomes for the host population, such as extinction, multiple
attractors, disease established or epidemic break out in periodic manner. Thus, system
(1.1) is very sensitive to perturbations and control methods. Here we give two phase
portraits of (1.1) with one and three endemic equilibria, respectively, to illustrate the
complication of dynamics (see Figs. 5, 6).

4 BT bifurcation and Hopf bifurcation of system (1.1)

In this section, we discuss if system (1.1) can undergo BT bifurcation and Hopf bifur-
cation in a small neighborhood of an endemic equilibrium. We will show that BT

123
Bifurcations of an epidemiological model 203

Fig. 6 The dynamics of system (1.1) with one endemic equilibrium, where r = 1, u = 0.35, d =
1.0813, α = 1.441 and σ = 6.5899. The endemic equilibrium is approximately (0.555, 0.035), which
is represented by ∗

bifurcation may occur for system (1.1) in a small neighborhood of an endemic equi-
librium which is a tangent point of L and C, and Hopf bifurcation may occur for system
(1.1) in a small neighborhood of an endemic equilibrium which is a cross point of L
and C.
We first study BT bifurcation of system (1.1). From Theorem 2.3, we know that
endemic equilibrium E(x1 , y1 ) is degenerate if it is a tangent point of L and C. In
Sect. 3 we have shown that this equilibrium is a degenerate elliptic equilibrium of
codimension 3 if it is a triple tangent point of L and C and it satisfies some condi-
tions. In the following we will show that under some conditions endemic equilibrium
E(x1 , y1 ) is a cusp of codimension 2 if it is a double tangent point of L and C.

Lemma 4.1 Suppose that system (1.1) has an endemic equilibrium E(x1 , y1 ) which
is a double tangent point of L and C and σ f (x1 ) − α f  (x1 ) = 0. Then the endemic
equilibrium E(x1 , y1 ) is a cusp of codimension 2 if E(x1 , y1 ) ∈ γ2 , where γ2 be the
part of the curve C with x ∗ = u+13 < x < 1 (see Fig. 1).

Proof Since E(x1 , y1 ) is a double tangent point of L and C, and σ f (x1 )−α f  (x1 ) = 0,
we obtain that x1 = PT + σα , y1 = α(σσ−1)2 , and

α u+1 rσ2  α α  α 

PT + = , 1 − σ = 2 PT + PT + − u PT + − 1 .
σ 3 α σ σ σ

We now move the endemic equilibrium E(x1 , y1 ) to the origin by setting z 1 =

x − x1 , z 2 = y − y1 . Then system (1.1) becomes

α(σ −1)
ż 1 = σ z 1 − αz 2 + 3r (− σα + u+1
3 − PT )z 1 − r z 1 ,
2 3
α(σ −1)2 α(σ −1) (4.1)
ż 2 = σ2
z 1 − σ z 2 − σ z 22 + (σ − 1)z 1 z 2 .

123
204 L. Cai et al.

α(σ −1)
Making a linear transformation X = z 1 , Y = σ z1 − αz 2 , system (4.1) is
converted to

Ẋ = Y + 3r (− σα + u+1
3 − PT )X − r X ,
2 3
α σ 2 αr (σ −1) 3
σ (σ − 1)(− σ + 3 − PT )X − (σ − 1)X Y + α Y −
Ẏ = 3αr u+1 2
σ X .
(4.2)

In a small neighborhood of the origin, we make a near identity smooth changes

 
σ 2 α u+1 σ
x=X− X , y = Y + 3r − + − PT X 2 − X Y.
2α σ 3 α

Then system (4.2) can be written as

ẋ = y + O(|x, y|3 ), ẏ = d1 x 2 + d2 x y + O(|x, y|3 ), (4.3)

where d1 = 3αr (σσ

−1)
(− σα + u+1 α
3 − PT ) and d2 = −(σ − 1) + 6r (− σ + 3 − PT ).
u+1
α
Since E(x1 , y1 ) ∈ γ2 , that is 3 < x1 = σ + PT , one has d1 < 0 and d2 < 0.
u+1

Thus, E(x1 , y1 ) is a cusp of codimension 2. We finish the proof.

We give an example to show that system (1.1) have a unique endemic equilibrium
and in a small neighborhood of this endemic equilibrium, system (1.1) can undergo
BT bifurcation.

Theorem 4.1 If (2.14) holds (i.e. PT = u and σ σ−1 = 4α r

), then system (1.1) has a
unique endemic equilibrium E A1 = ( 2 , 8α (1 − 2u)) and three disease-free equilib-
1 r

ria: O(0, 0) is a stable node, A(u, 0) is a saddle-node, and E 0 (1, 0) is a saddle. The
endemic equilibrium E A1 is a cusp of codimension 2 if

 
r 1 r + 2 − 4u (8u 2 − 1)r − 2(2u − 1)2
(α, σ, d) = + − u, , ,
4 2 2 − 4u 4(1 − 2u)

otherwise E A1 is a saddle-node. The global dynamics of system (1.1) with a cusp

endemic equilibrium is shown in Fig. 7.
If we choose (α, σ ) as bifurcation parameters, then system (1.1) undergoes BT
bifurcation in a small neighborhood of endemic equilibrium E A1 as (α, σ ) varies
near ( r4 + 21 − u, r +2−4u
2−4u ). Hence, there exist some parameter values such that system
(1.1) has a unstable limit cycle surrounding an endemic equilibrium, and there exist
some other parameter values such that system (1.1) has a unstable homoclinic loop.

Proof The first conclusion of the theorem comes from Lemmas 2.2, 2.3 and 4.1 by
straighten computation. And the second conclusion of the theorem can be proved by
normal form theory. Set

123
Bifurcations of an epidemiological model 205

Fig. 7 The phase portrait of system (4.4) at (λ1 , λ2 ) = (0, 0), where r = 4, u = 0.4, d = 1.3, α = 1.1
and σ = 11. The endemic equilibrium is approximately (0.5, 0.09), which is represented by ∗

⎧ ⎫
⎪
⎪ α = r4 + 21 − u, ⎪
⎪
⎪
⎨ (8u 2 −1)r −2(2u−1)2
⎪
⎬
(α, d, r, u, σ ) : d = > 0,
BT = 4(1−2u)
+2−4u
,
⎪
⎪ σ = 2−4u > 1,
r ⎪
⎪
⎪
⎩ ⎪
⎭
r > 0, 21 > u > 0

which forms a hypersurface in five-dimensional parameter space (α, d, r, u, σ ).

In a small neighborhood of the endemic equilibrium E A1 , we consider a perturba-
tion system of (1.1)

dx
dt = r (1 − x)(x − u)x − (α0 + λ1 )y,
dy (4.4)
dt = (−(α0 + λ1 ) − d0 − r u + (σ0 + λ2 − 1)x − (σ0 + λ2 )y)y,

where (α0 , σ0 , d0 ) ∈ BT , λ1 and λ2 are very small parameters. When λ1 = λ2 = 0,

system (4.4) has a cusp of codimension 2 at E A1 (see Fig. 7).
Following the process of deriving normal form, system (4.4) can be reduced to
⎧ dx
⎨ ds = y,
⎪ √
dy −8u+3)
2(8u 2√
= ν1 (λ1 , λ2 ) + ν2 (λ1 , λ2 )y + x 2 − ( 2(1−2u) + O(λ))x y (4.5)
⎪
⎩
ds 1−2u
+R(x, y, λ),

where s = −kt, k > 0, μi (·), O(·) and R(·) are smooth functions with respective to
their arguments. O(0) = 0, R(x, y, λ) have Taylor expansions in (x, y) starting with
at least cubic terms, and the Jacobian of (ν1 , ν2 ) with respect to (λ1 , λ2 ) at the origin
is
 
 ∂(ν1 ,ν2 )  +3(1−2u)2 (8u 2 −8u+1))
Jν =  ∂(λ ,λ )  = 4((4u−3)(5u−1)(4u−1)r
r 3 (r +2−4u)(1−2u)3
.
1 2 (0,0)

123
206 L. Cai et al.

Fig. 8 The BT bifurcation

diagram of system (4.4), where
λ1 and λ2 are bifurcation
parameters, S N ± are
saddle-node bifurcations, H is
Hopf bifurcation and H L is
homoclinic bifurcation

Therefore, there are values of (r, u), for instance, (4u − 3)(5u − 1)(4u − 1)r +
3(1 − 2u)2 (8u 2 − 8u + 1) = 0, such that Jν = 0. From the classic results in Bogdanov
(1981) and Takens (1974), system (4.5) undergoes BT bifurcation in a small neigh-
borhood of endemic equilibrium E A1 as (λ1 , λ2 ) varies in a small neighborhood of
the origin. Therefore, there exist some parameter values of (λ1 , λ2 ) such that system
(4.5) has a unstable limit cycle surrounding an endemic equilibrium, and there exist
some other parameter values such that system (4.5) has a unstable homoclinic loop.
This leads to the second conclusion of the theorem.
From Theorem 4.1, we can see that almost all orbits of system (4.4) go to the origin
if λ1 = λ2 = 0, see Fig. 7. Hence, the disease can not establish and almost all host
population goes to extinction even though R0 > 1 and there exists an endemic equi-
librium. However, as λ1 and λ2 vary in a small neighborhood of the origin (see Fig. 8),
system (4.4) has different dynamical behaviors, for example, as λ1 and λ2 cross the
curve S N + , system (4.4) undergoes saddle-node bifurcation. When parameters λ1 and
λ2 lie the region I, system (4.4) has two endemic equilibria, one is a hyperbolic saddle
and the other is a unstable node or a focus. When parameters λ1 and λ2 cross the curve
H into the region II (i.e., the region between H and H), the unstable focus becomes
stable, so an unstable limit cycle appears. When parameters λ1 and λ2 lie on the curve
H L,the limit cycle meet the saddle, change into homoclini loop. And as parameters λ1
and λ2 cross the curve H into the region III, system (4.4) has two endemic equilibria,
one is a hyperbolic saddle and the other is a stable node or a focus (see Fig. 9). Last
when parameters λ1 and λ2 cross the curve S N − into the region IV, system (4.4) has
no endemic equilibria. Biologically, this conclusion shows that bi-stable phenomenon
occurs and the disease can establish in the host population as (λ1 , λ2 ) are in the range
I ∪ I I ∪ I I I.
We now discuss Hopf bifurcation of system (1.1). From Theorem 2.3, we know
that endemic equilibrium E(x0 , y0 ) may be a weak focus if it is a cross point of L and
C, and it satisfies that σ σ−1 − f  (x0 ) > 0 and σ f (x0 ) − α f  (x0 ) = 0. In the general
case we first give the existence of weak focus of system (1.1).
Lemma 4.2 Suppose that system (1.1) has an endemic equilibrium E(x0 , y0 ) which
satisfies σ σ−1 − f  (x0 ) > 0 and σ f (x0 ) − α f  (x0 ) = 0. Then the endemic equilibrium

123
Bifurcations of an epidemiological model 207

Fig. 9 The phase portrait of system (4.4) at (λ1 , λ2 ) = (0.2, −0.3), where r = 4, u = 0.4, d = 1.3, α =
0.8 and σ = 11.2. The two endemic equilibria are approximately (0.5, 0.125) and (0.715, 0.321), which
are represented by ∗

E(x0 , y0 ) is a stable (respectively unstable) weak focus of order one if σ > σ1 (respec-
tively 1 <√σ < σ1 ), and E(x0 , y0 ) may be a weak focus of order two if σ = σ1 , where
B+ B 2 +12α(u+1)
σ1 = 2(u+1) , B = 2r + r u + 3d + 1 + u + 2r u 2 .

Proof Since σ f (x0 )−α f  (x0 ) = 0 and E(x0 , y0 ) is an endemic equilibrium of system
(1.1), we can find that
  
x0 = 6r 2r (u + 1) − (σ
1
− 1) + (2r (u + 1) − (σ − 1))2 + 12r (α + d) ,
σ −1
y0 = σ (x 0 − PT ) > 0.

We first move the endemic equilibrium E(x0 , y0 ) to the origin. Let z 1 = x−x0 , z 2 =
y − y0 . Then by σ f (x0 ) − α f  (x0 ) = 0, system (1.1) becomes

ż 1 = (σ − 1)(x0 − PT )z 1 − αz 2 + r (u + 1 − 3x0 )z 12 − r z 13 ,
(σ −1)2
(4.6)
ż 2 = σ (x0 − PT )z 1 − (σ − 1)(x0 − PT )z 2 + (σ − 1)z 1 z 2 − σ z 22 .

σ −1
Note that σ − 1 > 0, x0 − PT > 0 and σ − f  (x0 ) > 0. This leads to

β = σ (x0 − PT ) (α − σ x0 + σ PT ) > 0. (4.7)

We make the following linear transformation of state variables and time variable,
respectively,
   (x0 −P
  √
z1 1 √ T )σ x β (σ − 1)
β
= (x0 −P√
T )(σ −1)
, τ= t.
z2 0 β
y σ

123
208 L. Cai et al.

Then system (4.6) becomes

 dx
dτ = −y + a20 x 2 + a02 y 2 + a11 x y + a30 x 3 + a03 y 3 + a21 x 2 y + a12 x y 2 ,
dy
(4.8)
dτ = x + b11 x y,

where
√
σ (1+2r +2r u−σ −6r x0 ) r σ (u+1−3x0 ) β r σ (u+1−3x
√ 0) ;
a11 = (σ −1)(α−σ x√ 0 +σ PT )
; a02 = ; a20 = (σ −1) β
(σ −1)(α−σ x0 +σ PT )2
3r σ β 3r σ rσ √
a12 = − ; a21 = − (σ −1)(α−σ x0 +σ PT ) ; a30 = − (σ −1) β
;
(σ −1)(α−σ x0 +σ PT )2
r σ (x0 −PT )
2
√σ .
a03 = − (σ −1)(α−σ x0 +σ PT )2
; b11 = β

We finally determine if the equilibrium (0, 0) of system (4.8) is a weak focus by suc-
cessor function. It is convenient to introduce polar coordinates (ρ, θ ) and to rewrite
system (4.8) in polar coordinates by x = ρ cos θ, y = ρ sin θ . It is clear that in a
small neighborhood of the origin the successor function D(c0 ) of system (4.8) can be
expressed by

D(c0 ) = ρ(2π, c0 ) − ρ(0, c0 ),

where ρ(θ, c0 ) is the solution of the following system

 dρ
dθ = R2 (θ )ρ 2 + R3 (θ )ρ 3 + R4 (θ )ρ 4 + R5 (θ )ρ 5 + · · · ,
ρ(0) = c0 , 0 < |c0 | 1,

where

R2 (θ ) = cos θ (a02 + b11 + a11 cos θ sin θ + cos2 θ (a20 − a02 − b11 )),
R3 (θ ) = cos θ (a11
2 cos2 θ sin θ − 2a 2 cos2 θ sin θ + 2a a cos5 θ +
02 02 11
a20 b11 cos θ sin θ − 2a20 a02 cos4 θ sin θ + 2a02 b11 cos4 θ sin θ + a02
2 2 sin θ

+2a20 a02 cos θ sin θ − 2a20 b11 cos θ sin θ − 3a02 b11 cos θ sin θ
2 4 2

+a12 cos θ − 2a20 a11 cos5 θ + b11 2 cos4 θ sin θ + b a sin θ + 2a a cos3 θ
11 02 20 11
+a03 sin θ + 2a02 a11 cos θ + a20 cos4 θ sin θ − b11
2 2 cos2 θ sin θ − 3b a cos3 θ
11 11
−a03 sin θ cos2 θ − a112 cos4 θ sin θ + 2b a cos5 θ + a cos2 θ sin θ
11 11 21
b11 a11 cos θ − a12 cos3 θ − 4a02 a11 cos3 θ + a022 cos4 θ sin θ + a cos3 θ ),
30

and Ri (θ ) is a polynomial of (sin θ, cos θ ), i = 4, 5, . . ., whose coefficients can be

expressed by the coefficients of system (4.8). We omit them here since the expressions
are too long.
From the method of successor function in Andronov et al. (1971) and Zhang et al.
(1992), we can obtain the first Liapunov number of the equilibrium (0, 0) of system
(4.8)

r ασ S
C1 = 1
8 (a30 + a12 + a11 (a20 + a02 )) = √ ,
8(σ −1)2 (α−σ x0 +σ PT ) β

123
Bifurcations of an epidemiological model 209

where S = − (u + 1) σ 2 + 2r + r u + 3d + 1 + u + 2r u 2 σ + 3α and β is as in
(4.7).
Obviously, C1 has the same sign of S. Let us discuss the sign of S. Set B =
2r + r u + 3d + 1 + u + 2r u 2 . Then

S = −(u + 1)(σ − σ1 )(σ − σ2 ),

√ √
B+ B 2 +12α(u+1) B− B 2 +12α(u+1)
where σ1 = 2(u+1) > 1 and σ 2 = 2(u+1) < 0.
Therefore, the endemic equilibrium E(x0 , y0 ) is a unstable (stable) weak focus of
order one if 1 < σ < σ1 (σ > σ1 , respectively).
However, if σ = σ1 , then C1 = 0. We further compute the second Liapunov number
of equilibrium (0, 0) of system (4.8) as C1 = 0 and obtain
1 1 1 1
C3 = a11 a02 a03 + a02 a21 a11 + a20 a11 a21 + a12 a21
6 24 16 48
1 1 1 2 1
+ a12 a11 + b11 a11 a20 a02 + a02 a11 a20 + b11 a11 a03
2
72 16 6 24
7 1 1 2 1
+ a20 a11 a03 + a12 b11 a02 + a20 a02 a11 + a12 b11 a20
48 16 16 16
1 1 1 5
+ b11 a11 a02 + a12 a20 a02 + a02 a11 + a12 a02
2 3 2
16 6 72 48
1 1 5 3 1
+ a12 a03 + a12 a20 2
+ a02 a11 + a20 a113
,
16 16 48 72
which can be expressed by parameters (r, u, α, d) of system (4.8). However, it is very
long so we omit it.
If C3 = 0, then endemic equilibrium E(x0 , y0 ) is a weak focus of order two, and
endemic equilibrium E(x0 , y0 ) is stable (unstable) as C3 < 0 (C3 > 0, respectively).
Thus, we complete the proof.

We now provide an example to show that system (1.1) has an endemic equilibrium
which is a weak focus of order two. And under a small perturbation, the system (1.1)
can undergo Hopf bifurcation and degenerate Hopf bifurcation, and produce at least
two limit cycles.

Theorem 4.2 Suppose that (H 2) holds. Then in the parameters space (α, d, r, u, σ )
there exists a hypersurface H0 and a surface H1 ⊂ H0 , where

⎧ ⎫
⎪ α = 29 σ (1 + 2u)(1 − u), ⎪
⎪
⎪ ⎪
⎪
⎪
⎨ d = 19 (1 + 2u)((2u + 1)σ − 3r − 3), ⎪
⎪
⎪
⎬
H0 = (α, d, r, u, σ ) : 8(1−u)(1+2u)(σ −1)
> r > 0, σ > 1, ,
⎪
⎪ 3(4−4u+3u 2 ) ⎪
⎪
⎪
⎪ √ 2 √ 1 ⎪
⎪
⎪
⎩ (37+9 17) 3 −2−2(37+9 17) 3 ⎪
⎭
2 >u >
1
√ 1
6(37+9 17) 3

123
210 L. Cai et al.

⎧ ⎫
⎪ α = 29 σ (1 + 2u)(1 − u), ⎪
⎪
⎪ 2 +2u−1)σ +2u−1−10u 2 ) ⎪⎪
⎪
⎪ d = (1+2u)((4u +4u
3
, ⎪
⎪
⎨ 9(2u 2 −u+1) ⎬
H1 = (α, d, r, u, σ ) : r = (2−u−4u )(σ −1)
2
, σ > 1, ,
⎪
⎪ 3(1−u+2u 2 ) ⎪
⎪
⎪
⎪ √ 2 √ 1 ⎪
⎪
⎪
⎩ 1
> u > (37+9 17) −2−2(37+9 17)
3 3 ⎪
⎭
2 √ 1
6(37+9 17) 3

such that system (1.1) has a unique endemic equilibrium E(x0 , y0 ) as (α, d, r, u, σ ) ∈
r (1−u)
H0 , where x0 = 1+2u 3 , y0 = 3σ . And E(x 0 , y0 ) is a weak focus of order one
if (α, d, r, u, σ ) ∈ H0 \H1 , and E(x0 , y0 ) is a unstable weak focus of order two if
(α, d, r, u, σ ) ∈ H1 .
Moreover, as parameters (α, d, r, u, σ ) vary in the hypersurface H0 and pass
through the surface H1 , system (1.1) undergoes degenerate Hopf bifurcation and
produce a unstable limit cycle near E(x0 , y0 ). And as parameters (α, d, r, u, σ ) fur-
ther vary in the parameters space and pass through the hypersurface H0 , system (1.1)
undergoes Hopf bifurcation again and produce a stable limit cycle near E(x0 , y0 ).
Therefore, there are parameters values such that system (1.1) has two limit cycles, the
attracting cycle is surrounded by a repelling cycle.

Proof If (α, d, r, u, σ ) ∈ H0 , then system (1.1) becomes

 dx 9r x(1−x)(x−u)
= − αy,
dt 
2σ (1+2u)(1−u)  (4.9)
dy 3(σ −1)x+r +1−σ −2uσ −r u+2u
dt = σy 3σ −y .

8(1−u)(1+2u)(σ −1)
Since 0 < r < 3(4−4u+3u 2 )
, by calculation we know that


9r x(1−x)(x−u)
2σ (1+2u)(1−u) − αy = 0, (4.10)
3(σ −1)x+r +1−σ −2uσ −r u+2u
3σ −y=0

has only a pair of real simple root (x0 , y0 ). Thus, system (4.9) has a unique endemic
r (1−u)
equilibrium E(x0 , y0 ), where x0 = 1+2u 3 , y0 = 3σ .
If (α, d, r, u, σ ) is in a small neighborhood of H0 , then system (4.9) has still a
unique endemic equilibrium by the continuity of roots with respect to coefficients.
We now study the local dynamics of equilibrium E(x0 , y0 ) if (α, d, r, u, σ ) ∈ H0 .
Moving E(x0 , y0 ) to the origin, we obtain the following system

 dx
dt = 13 r (1 − u)x − 29 σ (1 + 2u)(1 − u)y − r ux 2 − r x 3 ,
(4.11)
dy
dt = 3σ (σ
r
− 1)(1 − u)x − 13 r (1 − u)y + (σ − 1)x y − σ y 2 .

It is clear that the eigenvalues of system (4.11) at the origin are a pair of purely
imaginary numbers. Therefore, E(x0 , y0 ) is a center or weak focus.

123
Bifurcations of an epidemiological model 211

We further claim that it is a weak focus. We first make linear transformations of

state variables (x, y) and time variable t for system (4.11) respectively,
⎛  ⎞
  1 3r   
x β̃ X 1−u
=⎝ 
σ −1 3r
⎠ , τ= 3r β̃ t,
y 0 Y 9
σ β̃

where we have denoted β̃ = 2σ − 2 − 3r + 4uσ − 4u to simplify notations. We obtain

the following system
⎧ dX
⎨ dτ = −Y + a20 X 2 + a11 X Y + a02 Y 2 + a30 X 3 + a21 X 2 Y
+a12 X Y 2 + a03 Y 3 , (4.12)
⎩ dY
dτ = X + b11 X Y,

where

9(2r u+σ −1) 3ra20 3ra11
a20 = 3u 3r
, a11 = , a02 = , a21 = 2r u+σ −1 ,
u−1 β̃ (u−1)β̃ β̃
a12 = 9ra20
, a03 = ra21
, a30 = a20
u , b11 = − a20 (σ −1)
.
u β̃ β̃ ru

By the method of successor function, we let X = ρ cos θ, Y = ρ sin θ. Then in a

small neighborhood of the origin, system (4.12) can be modified to

dρ
= R2 (θ )ρ 2 + R3 (θ )ρ 3 + R4 (θ )ρ 4 + R5 (θ )ρ 5 + O(ρ 6 ), (4.13)
dθ

where Ri (θ ) is a polynomial of (sin θ, cos θ ), for i = 1, 2, . . ..

From Lemma 4.2, we obtain the first Liapunov number of equilibrium (0, 0) of
system (4.12) to be,

9 3r β̃(σ −1)(1+2u)(4u 2 σ +6r u 2 − 4u 2 + uσ − u − 3r u − 2σ + 2 + 3r )
C1 = .
(1 − u)2 β̃ 3

(2−u−4u 2 )(σ −1) (2−u−4u 2 )(σ −1)

It is clear that C1 > 0 if r > 3(1−u+2u 2 )
, C1 = 0 if r = 3(1−u+2u 2 )
and
(2−u−4u 2 )(σ −1)
C1 < 0 if r < 3(1−u+2u 2 )
. Hence, the equilibrium E(x0 , y0 ) is a unstable weak
focus of order one if the parameters (α, d, r, u, σ ) ∈ H0+ , where
⎧ ⎫
⎪
⎪ α = 29 σ (1 + 2u)(1 − u), ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ d = 19 (1 + 2u)((2u + 1)σ − 3r − 3), ⎪
⎪
⎨ ⎬
H0+ = (α, d, r, u, σ ) : 8(1−u)(1+2u)(σ2 −1) > r > (2−u−4u )(σ2−1) ,
2
,
⎪
⎪ 3(4−4u+3u ) 3(1−u+2u ) ⎪
⎪
⎪
⎪ √ √ ⎪
⎪
⎪
⎪
2 1
⎪
⎪
⎩ σ > 1, 21 > u > (37+9 17) −2−2(37+9 17) 3 ⎭
3
√ 1
6(37+9 17) 3

123
212 L. Cai et al.

and the equilibrium E(x0 , y0 ) is a stable weak focus of order one if the parameters
(α, d, r, u, σ ) ∈ H0− , where

⎧ ⎫
⎪
⎪ α = 29 σ (1 + 2u)(1 − u), ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎨ d = 9 (1 + 2u)((2u + 1)σ − 3r − 3), ⎪
1 ⎬
−
H0 = (α, d, r, u, σ ) : (2−u−4u 2 )(σ −1) > r > 0, σ > 1, .
⎪
⎪ 3(1−u+2u 2 ) ⎪
⎪
⎪
⎪ √ √ ⎪
⎪
⎪ (37+9 17) 3 −2−2(37+9 17) 3 ⎪
2 1
⎪ ⎪
2 >u >
1
⎩ √ 1 ⎭
6(37+9 17) 3

However, C1 = 0 in the case r = (2−u−4u )(σ −1)

2
3(1−u+2u 2 )
, which is equivalent to that
(α, d, r, u, σ ) ∈ H1 . To determine the dynamics of the equilibrium E(x0 , y0 ), we
have to compute the second Liapunov number
√ of equilibrium (0, 0) of system (4.12)
9s1 (1+2u)(1−u+2u 2 )3 u(3+4u+8u 2 )(2−u−4u 2 )
and we obtain C3 = 4u 4 (1−u+2u 2 )(1−u)2 (3+4u+8u 2 )4
, where

s1 = 40u 3 + 54u 2 − 3u − 1.

√ 2 √ 1
(37+9 17) 3 −2−2(37+9 17) 3
Since 1
2 >u> √ 1 , one has s1 > 0. Consequently, C3 > 0 if
6(37+9 17) 3
(α, d, r, u, σ ) ∈ H1 . So the equilibrium E(x0 , y0 ) is a unstable weak focus of order
two if (α, d, r, u, σ ) ∈ H1 . Thus, we complete the proof of the first conclusion.
Let us consider the Hopf bifurcation of system (4.9). We fix three parameters
(d, σ, u) and choose (α, r ) as bifurcation parameters. As parameters (α, r ) vary in the
hypersurface H0 and parameter r pass through the surface H1 from H0+ to H0− , equi-
librium E(x0 , y0 ) changes the stability from unstable weak focus of order two to stable
weak focus of order one, and it is easy to check that the transversal condition holds.
Hence, system (4.9) undergoes degenerate Hopf bifurcation and produce a unstable
limit cycle. As parameters (α, r ) vary in the parameters space and parameter α passes
through the hypersurface H0 from α < 29 σ (1 + 2u)(1 − u) to α > 29 σ (1 + 2u)(1 − u),
the equilibrium E(x0 , y0 ) changes the stability again from stable weak focus of order
one to unstable hyperbolic focus and the transversal condition of eigenvalues with
respect to α holds. Thus, system (1.1) undergoes Hopf bifurcation and produce a sta-
ble limit cycle. Hence, in parameters space (α, d, r, u, σ ), H0 is the Hopf bifurcation
hypersurface and H1 is the degenerate Hopf bifurcation surface.
Therefore, there exist values of the parameters such that system (1.1) has a unique
endemic equilibrium and two limit cycles surrounding this equilibrium (see Fig. 10).

From Theorem 4.2, we can see that system (1.1) has a unique endemic equilibrium,
a stable limit cycle and a unstable limit cycle for some values of parameters. Thus,
tristable phenomenon occurs and epidemic can be observed to break out in a periodic
manner for some initial values of the host population.

123
Bifurcations of an epidemiological model 213

Fig. 10 System (1.1) has two limit cycles, where r = 0.2582897033, u = 0.45, d = 0.0053054102, α =
0.4644444444 and σ = 2. The endemic equilibrium is approximately (0.63333, 0.02367), which is repre-
sented by ∗

5 Discussion and conclusion

In this paper we completely study the bifurcations and dynamical behavior of an epi-
demiological model with strong Allee effect. The qualitative conclusions presented for
the model support the numerical bifurcation analyses and conjectures in Hilker et al.
(2009). Hence, we highlight the impact of both disease and strong Allee effect on
population persistence observed in Hilker et al. (2009). Moreover, we gain new bifur-
cations phenomena such as pitchfork bifurcation, BT bifurcation of codimension two,
degenerate Hopf bifurcation and degenerated BT bifurcation of codimension three
in elliptic case. These rich bifurcations exhibit complicated dynamical behavior of
the model, for example, the multiple attractors, homoclinic loop, one limit cycle, two
limit cycles and etc.. When parameters (α, d, r, u, σ ) vary in a small neighborhood of
pitchfork bifurcation surface, model (1.1) has either an endemic equilibrium or three
endemic equilibria. Hence, in this case the disease can establish in the host popula-
tion. When parameters (α, d, r, u, σ ) vary in a small neighborhood of BT bifurcation
surface, model (1.1) has two endemic equilibria and a unstable limit cycle or homo-
clinic loop for corresponding values of parameters, which implies that one endemic
equilibrium becomes an attractor. Thus, the infected population will tend to a constant
for the initial number of host populations in an open set, and sustained oscillation can
not be observed. And when parameters (α, d, r, u, σ ) vary in a small neighborhood
of degenerate Hopf bifurcation surface, model (1.1) has two limit cycles for some
values of parameters, one of which is stable. This leads that epidemic can be observed
to break out in a rather reproducible periodic manner in the host population. These
conclusions reveal that the dynamics of model (1.1) is very sensitive to parameters
perturbation. Hence, biologically our results have important consequences for disease
control and biological conservation. And it also provides, at least in theory, a key to
understanding the impact of diseases and Allee effect for the threatened populations

123
214 L. Cai et al.

or the success of re-introductions, which may help us to prevent both epidemics and
Allee effects driven extinctions.
Mathematically, in this paper we provide an approach to seek possible equilibria at
which bifurcations occur. These equilibria are tangent points of two nullclines for two
dimensional mathematical models. Moreover, we give explicit smooth transformations
to reduce a system with an nilpotent equilibrium into a normal form of codimension
3, which is very useful for analysis of degenerate BT bifurcation. And the method to
study degenerate Hopf bifurcation of codimension 2 of system (1.1) has its generality,
which can also be applied to investigate the other related mathematical models. There-
fore, the methods developed in our paper are some approaches to study bifurcations
and dynamics of two dimensional mathematical models with multi-parameters.

Acknowledgments We are very grateful to Prof. Mark Lewis and the two anonymous referees for their
valuable comments and suggestions, which led to an improvement of our original manuscript.

References

Allee WC (1931) Animal aggregation: a study in general sociology. University of Chicago Press, Chicago
Anderson RM, May RM (1979) Population biology of infectious diseases. Nature 280:361–367
Andronov A, Leontovich AE, Gordon II, Maier GA (1971) Theory of bifurcations of dynamical systems
on a plane. Israel Program for Scientific Translations, Jerusalem
Angulo E, Roemer GW, Berec L, Gascoigne J, Courchamp F (2007) Double Allee effects and extinction
in the island fox. Conserv Biol 21:1082–1091
Baer SM, Kooi BW, Kuznetsov Yu A, Thieme HR (2006) Multiparametric bifurcation analysis of a basic
two-stage population model. SIAM J Appl Math 66:1339–1365
Bogdanov IR (1981) Versal deformation of a singularity of a vector field on the plane in the case of zero
eigenvalues. Selecta Math Sov 1:389–421
Chow S-N, Hale JK (1982) Methods of bifurcation theory. Springer, New York
Chow S-N, Li CZ, Wang D (1994) Normal forms and bifurcation of planar vector fields. Cambridge Uni-
versity Press, Cambridge
Clifford LD, Mazet JAK, Dubovi EJ, Garcelon DK, Coonan TJ, Conrad PA, Munson L (2006) Pathogen
exposure in endangered island fox (Urocyon littoralis) populations: implications for conservation
management. Biol Conserv 131:230–243
Courchamp F, Berec L, Gascoigne J (2008) Allee effects in ecology and conservation. Oxford University
Press, Oxford
Daszak P, Berger L, Cunningham AA et al (1999) Emerging infectious diseases and amphibian population
declines. Emerg Infect Dis 5:735–748
de Castro F, Bolker B (2005) Mechanisms of disease-induced extinction. Ecol Lett 8:117–126
Deredec A, Courchamp F (2006) Combined impacts of Allee effects and parasitism. Oikos 112:667–679
Dumortier F, Fiddelaers P, Li C (2001) Generic unfolding of the nilpotent saddle of codimension four. In:
Global analysis of dynamical systems. Institute of Physics, Bristol, pp 131–166
Dumortier F, Roussarie R, Sotomayor S (1987) Generic 3-parameter families of vector fields on the plane,
unfolding a singularity with nilpotent linear part. The cusp case of codimension 3. Ergodic Theory
Dyn Syst 7:375–413
Dumortier F, Roussarie R, Sotomayor S, Zoladek H (1991) Bifurcations of planar vector fields Nilpotent
singularities and Abelian integrals, Lecture Notes in Mathematics, 1480. Springer-Verlag, Berlin
Etoua R, Rousseau C (2010) Bifurcation analysis of a generalized Gause model with prey harvesting and
a generalized Holling response function of type III. J Differ Equ 249:2316–2356
Friedman A, Yakubu AA (2011) Fatal disease and demographic Allee effect: population persistence and
extinction. J Biol Dyn. doi:10.1080/17513758.2011.630489
Hadeler PK, van den Driessche P (1997) Backward bifurcation in epidemic control. Math Biosci 146:15–35

123
Bifurcations of an epidemiological model 215

Haydon DT, Laurenson MK, Sillero-Zubiri C (2002) Integrating epidemiology into population viability
analysis: managing the risk posed by rabies and canine distemper to the Ethiopian wolf. Conserv Biol
16:1372–1385
Hilker FM, Langlais M, Petrovskii SV, Malchow H (2007) A diffusive SI model with Allee effect and
application to FIV. Math Biosci 206:61–80
Hilker FM, Lewis MA, Seno H et al (2005) Pathogens can slow down or reverse invasion fronts of their
hosts. Biol Invasions 7:817–832
Hilker FM, Langlais M, Malchow H (2009) The Allee effect and infectous diseases: extinction, multista-
bility and the (dis-)appearance of oscillations. Am Nat 173:72–88
Hudson PJ, Rizzoli AP, Grenfell BT et al (2001) The ecology of wildlife diseases. Oxford University Press,
Oxford, pp 45–62
Kuznetsov Yu A (1998) Elements of applied bifurcation theory, 2nd edn. In: Applications of mathematical
sciences, vol 112. Springer, New York
Lamontagne Y, Coutu C, Rousseau C (2008) Bifurcation analysis of a predator–prey system with general-
ised Holling type III functional response. J Dyn Differ Equ 20:535–571
Li C, Rousseau C (1989) A system with three limit cycles appearing in a Hopf bifurcation and dying in a
homoclinic bifurcation: the cusp of order 4. J Differ Equ 79:132–167
Mardešić P (1992) Versal unfolding of the cusp of order n. CR Acad Sci Paris Sér I Math 315:1235–1239
Perko L (1996) Differential equations and dynamical systems. Springer, New York
Takens F (1974) Singularities of vector fields. Publ Math IHES 43:47–100
Thieme HR, Dhirasakdanon T, Han Z, Trevino R (2009) Species decline and extinction: synergy of infec-
tious diseases and Allee effect? J Biol Dyn 3:305–323
Xiao D (1993) Bifurcations of saddle singularity of codimension three of a planar vector field with nilpotent
linear part Sci Sin A 23:252–260
Xiao D (2008) Bifurcations on a five-parameter family of planar vector field J Dyn Differ Equ 20:961–980
Xiao D (2005) Jennings LS Bifurcations of a ratio-dependent predator-prey system with constant rate har-
vesting SIAM J Appl Math 65:737–753
Xiao D, Ruan S (1999) Bogdanov–Takens bifurcations in predator-prey systems with constant rate harvest-
ing Fields Inst Commun 21:493–506
Xiao D, Zhang KF (2007) Multiple bifurcations of a predator–prey system Disc Cont Dyn Syst Ser B
8:417–437
Zhang Z, Ding T, Huang W, Dong Z (1992) Qualitative theory of differential equations. Science Press,
Beijing (in Chinese) (English edition: Qualitative theory of differential equations, translations of
mathematical monographs, vol 101. American Mathematical Society, Providence

123