Professional Documents
Culture Documents
IN EPIDEMIC MODELS
LOURDES ESTEVA
DEPARTAMENTO DE MATEMATICAS
Stability in nonlinear population models is often established by examining the eigenvalues of the linearized dynamics
This methods gives only stability relative to infinitesimal perturbations of the initial state.
Populations on the real word are subjected to large
perturbations.
It is essential that a population model should be stable relative to finite perturbations of its initial state.
In this course we deal with global properties of classic
epidemic models using the direct Lyapunov method and
topological approaches.
SI
S I R
S
I
R
S 0(t) = SI S
I 0(t) = SI I
R0(t) = I R
(1)
(2)
EQUILIBRIUM POINTS
Disease free equilibrium E0 = (1, 0).
Endemic equilibrium E1 = (S , I )
1
, I = (R0 1),
S =
R0
R0 =
.
+
Basic reproduction number: Average number of secondary infections of an infective during the infection period.
LINEAR ANALYSIS
Characteristic roots of the linearization around E0: , (+
)(R0 1).
E0 is a stable node if R0 < 1 and a saddle if R0 > 1.
Characteristic equation of the linearization around E1:
s2 + R0 s + (R0 1).
Routh Hurwitz criteria (Gantmacher, 1959) E1 is l.a.s
for R0 > 1.
5
GLOBAL STABILITY
Poincar`
e-Bendixon Theorem (Coddington & Levinson, 1955).
For two dimensional systems, bounded paths approach
a) an equilibrium point,
b) a limit cycle or,
c) a cycle graph.
Limit cycles must contain alt least one equilibrium in
their interior.
Cyclic graphics are not possible from a stable equilibrium.
Bendixon-Dulac test (Hethcote, 1976).
x0(t) = F (x, y)
y 0(t) = G(x, y)
(x, y) D simply connected, F (x, y), G(x, y) C 1(D)
(HF )/x + (HG)/y sign stable in D for some
H(x, y) C 1 (D)
there is no periodic solution or cyclic graphs in D.
6
R0 1:
E0 T is the only equilibrium point in T.
There is not limit cycle in T.
There is not cyclic graph in T.
All paths in T approach E0 .
R0 > 1:
E0 is a saddle, (S, 0) E0 for 0 S 1.
E1 int T is l.a.s.
H = 1/I
(S + (1 S)/I) + (S ) = /I
S
I
all paths in T except the S axis approach E1 .
V : U Rn R;
V C 1 (U)
is positive definite on U if
(i)V (0) = 0
(ii) V (
x) > 0, x 6= 0 U.
V is negative definite if V is positive definite.
x(t)0 = f (
x)
x = (x1 , x2, ...xn) Rn,
f (
x) = (f1 (
x), f2(
x), ..., fn(
x)) C 1 (Rn).
The orbital derivative of V along the trajectory x(t)
V (
x(t)) 0
xi(t)
V (
x(t)) = ni=1
xi
10
I
ln
+
W
I
I
V (S, I) = W1 S S S ln
2
S
I
for some W1 > 0, W2 > 0.
V = W1 (S S )(I + )
S
+ W2 (I I )(S ( + ))
From the equations at equilibrium:
= I , ( + ) = S
S
V = (W2 W1)(S S )(I I ) W1 (S S )2
W1 = W2 = 1
2
(S
S
)
V =
0
SS
V = 0 S = S M = {(S , I )}.
All paths in T+ approach E1. Since the vector field on
the I axis, and on the line S +I = 1 points to the interior
of T all paths in T {(S, 0)|0 S 1} tend to E1.
11
(3)
n
= {
z Rn|, zi 0, i = 1, ...n}
R+
(iv)
n
n
= {
z R+
|ni=1zi 1} or = {
z R+
|zi 1}
12
13
EXAMPLES
I) Kermack & McKendrick model
S 0(t) = SI S
I 0(t) = SI ( + )I
0
A=
, e =
, c =
,B = 0
0
( + )
0
II) Multi group SIS model (Lajmanovich & Yorke,
1979)
Si + Ii = 1, i = 1, ..n.
Ii0 = (1 Ii)nj=1 ij Ij (i + i)Ii
11 12
21
22
A=
n1 n2
c = 0,
1n
2n
nn
0 12
21 0
B=
n1 n2
14
11 (1 + 1)
( + )
22
2
2
e =
nn (n + n)
1n
2n
zi
V (
z ) = ni=1Wi zi zi ziln
zi
(4)
z 1) z0
V (
z ) = (
z z)T W A diag(
15
(5)
At the equilibrium z:
0 = diag(
z ) (
e + A
z ) + b(
z )
1
e = Az diag(
z ) b(z )
1
where z
Substituting e:
1
z ) B](
z z)
z 0(t) = diag(z) [A + diag(
1
diag(
z z) diag(
z )b(
z)
V becomes
b
b
(z)
(z)
1
n
,
..,
)
(
z
z
V (
z ) = (
z
z )T W A diag
z1z1
znzn
(6)
1
A = A + diag(
z ) B.
16
C a real n n matrix.
C Sw there exists a positive diagonal real matrix W such that W C + C T W is positive definite.
C is W skew-symmetrizable there exists a positive diagonal real matrix W such that W C is skewsymmetric (aji = aij , aii = 0).
If
bn(z)
b
(z)
1
A diag
, ..,
SW
z1z1
znzn
V (
z ) 0,
V (
z ) = 0 z = z
z is g.a.s within +.
17
If W A is skew-symmetric
Wibi(
z)
V (
z ) = ni=1
(zi zi)2 0.
zi zi
V (
z) = 0
z ) > 0}.
z R = {
z |zi = zi, i = 1, .., n s.t. bi(
Associate a graph to A in which nodes representing epidemiological classes zi, and arrows the mutual interactions following the rules
z ) = 0, zi is represented by i.
(i) If bi(
(ii) If bi(
z ) > 0, zi is represented by i.
(iii) Each pair of elements aij aj i < 0 is represented by
an arrow connecting nodes i and j.
18
Lemma 1. ( Cooke & Yorke, 1973). A W -skew symmetrizable n n matrix. If the associated graph is either
a) a tree and p 1 of the p terminal nodes are ,
b) or a chain and two consecutive internal nodes are ,
c) or a cycle and two consecutive nodes are ,
then M = {
z } within R.
1
.
&
3
a
1 2 3
b
1
%
&
I) SIR model
S 0(t) = SI S
I 0(t) = SI ( + )I
0
A = A =
, B = 0, b(
z ) = c =
0
0
S =
1 ( + )(R0 1)
,I =
, R0 =
R0
+ R0
+
20
S = S + /
( + )S
S0(t) = ( + )(1 + /) SI
( + + )I
I 0(t) = SI
A=
0
0
, e =
( + )
( + + )
B = 0.
A = A is skew-symmetric, b(
z ) = c.
21
, c =
( + )(1 + /)
0
11 12
21 22
, e =
B=
z = (I1 , I2)T ,
A=
11 (1 + 1)
22 ( + 2 )
0 12
21 0
b(
z ) = (12 I2, 21 I1)T
11
21 (1 I2)
I2
22
12 (1
I1
22
I1)
, c = 0,
To show:
12 I2 21 I1
C = A diag
,
SW .
I1I1 I2I2
12I2
12(1 I1)
W1
W1 11 + I I1
I1
1
WC =
21 I1
21 (1 I2 )
W2
W
+
2
22
I2
I2I2
Choosing W1 , W2:
12 (1 I1)
21 (1 I2)
W2 =
W1
I2
I1
C is symmetric
Det(C) > 0 C SW .
z R2|Ii 1, i = 1, 2}
z is g.a.s in = {
23
S1 + I1 = 1
S3 + I3 = 1
0 12 0
(1 + 1)
A = 21 0 23 , e = (2 + 2) , c = 0,
0 32 0
(3 + 3)
0 12 0
B = 21 0 23
0 32 0
24
12 (1
I1
I1)
21(1 I2)
(1
I
)
23
2
b(
0
z ) = B z, A =
I2
I2
32 (1 I3)
0
0
I3
b
b
b
(
z
)
(
z
)
(
z
)
1
2
3
C = W A diag
,
,
becomes
I1I1 I2 I2 I3I3
I1)
12 I2
12 (1
W
W1
0
1
I
I
I
1
1
1
21 (1 I2)
(
I
+
I
)
(1
I
)
21
1
23
3
23
2
W
W
W2
2
2
I
I
I
I
2
2
2
2
32 (1 I3)
32 I2
0
W3
W3
I3
I3 I3
is symmetric if
W1 > 0,
12(1 I1)I2
W2 =
W1 ,
21(1 I2)I1
25
23 (1 I2)I3
W3 =
W2 .
32 (1 I3)I2
26
For an epidemiological model with arbitrary n dissimilar groups is difficult to apply the above results.
Lajmanovich & Yorke (1976) proved existence and
stability of the endemic equilibrium for n multigrup
SIS model using Lyapunov functions and PerronFrobenius Theorem for positive and irreducible matrices.
Ii0 = (1Ii)nj=1 ij Ij (i +i)Ii,
Si +Ii = 1, i = 1, ..n
(7)
I0 = AI + N(I)
A=
(8)
11 (1 + 1)
12
21
22 (2 + 2)
n2
n1
=
N(I)
nj=11j Ij I1
nj=12j Ij I2
nj=1 nj Ij In
27
1n
2n
nn (n + n)
A=
(1 + 1)
12
21
(2 + 2)
Tr A = (1 + 1 + 2 + 2) < 0.
s(A) 0 det A = (1 + 1)(2 + 2) 12 21 0
21
12
1
(2 + 2) (1 + 1)
28
Proof of Theorem 3.
From Perron-Frobenius Theorem (Varga, 1962) it can be
proved the following
Lemma 2. (Lajmanovich & Yorke, 1976.) Let C an
irreducible n n matrix, and assume cij 0 whenever
i 6= j. Then, there exists an eigenvector w of C such that
w > 0 and the corresponding eigenvalue is s(A).
By Lemma 2, A has an eigenvector w > 0 with eigenvalue
s(A). Assume s(A) 0.
Define the Lyapunov function
= w I
V (I)
+ (w N(I))
= w I0(t) = (w AI)
V (I)
+ (w N(I))
= (AT w I)
+ (w N(I))
0
= s(A)(w I)
is the only invariant set contained in M = {I
{0}
= 0}
|V (I)
29
V s(A) ||w||||N(
I)||.
r = min wi
I = w I r||I||,
/r
||I||
Choose such that s(A) (||w||/r) > 0.
||I||
for ||I||
0 .
Choose 0 such that ||N(I)||
For [0, 0 ] :
||w||
s(A) ||w||
= s(A)
> 0.
V (I)
r
r
By Fixed Point Theorem 0 contains an endemic equilibrium I = (I1, .., In), Ii > 0.
Remark.
remains remains at a positive distance
If I 6= 0 then I(t)
from the boundary of for t 0.
30
Global stability of I
Remark. For V (
x) continuous Lyapunov Theorem is
still true replacing V (
x) by
V (
x(t + h)) V (
x(t))
.
V + = lim suph0+
h
Define M, m : R
= mini=1,..,n (Ii/I ).
m(I)
i
= maxi=1,..,n (Ii/I )
M(I)
i
Assume: M(I(t))
= I1(t)/I1 for [t0, t0 + ]
I 0(t0 )
M+(I(t0 )) =
I1
From (7):
0
I1 (t0 )
I1
I1(t0 )
= (1
Ij (t0 )I1
n
I1 (t0 ))j=1 1j
I1(t0 )
(1 + 1)I1
0 )) > 1:
If M(I(t
0
I1 (t0 )
I1
I1(t0 )
0 )) < 0
I10 (t0 ) < 0 M +(I(t
31
V (I)
0],
= max[1m(I),
0].
W (I)
0
0, W (I)
0 and V +(I)
V (I)
+(I)
0.
W
Ii I }
V = 0 in HV = {I|0
i
Ii 1} {0}
= 0 in HW = {I|I
W
i
By La Salle-Lyapunov Theorem solutions in approach
HV HW = {I} {0}.
6= 0 lim inft||I(t)||
> 0 then I is g.a.s.
Since I(0)
in {0}.
32
33
Nh0 = 0 Nh const.
Nv0 = Av Nv Nv
A
.
v
34
PARAMETERS
b biting rate. Average number of bites per mosquito
per day (once every two or three days).
h probability that an infectious bite produces a new
case in a susceptible human.
35
INFECTION RATES
Nh
probability that a mosquito chooses a human
Nh + m
as a host.
A human receives b
A mosquito takes b
Nv Nh
bites per day.
Nh Nh + m
Nh
human blood meals per day.
Nh + m
v b
Nh Ih
v b
Ih .
=
Nh + m Nh Nh + m
36
hb
Iv Sh h Sh
Nh + m
hb
Iv Sh (h + h )Ih
Nh + m
Rh0 (t) = h Ih h Rh
Sv0 (t) = A
Iv0 (t) =
(9)
v b
IhSv v Sv
Nh + m
v b
Ih Sv v Iv .
Nh + m
Sh + Ih + Rh = Nh
37
Sv + Iv = Nv
1. The subset
T : Sh + Ih + Rh = Nh ,
S v + Iv =
A
v
Sh
Ih
Rh
Sv
Ih
, ih =
, rh =
, sv =
, iv =
.
Nh
Nh
Nh
A/v
A/v
38
s0h(t) = h
hbA/v
iv sh h sh
Nh + m
i0h(t) =
hbA/v
iv sh (h + h)ih
Nh + m
i0v (t) =
v bNh
ih(1 sv ) v iv .
Nh + m
(10)
39
EQUILIBRIUM POINTS
Define
hv b2Nh A/v
R0 =
(Nh + m)2 v (h + h)
(11)
v bNh
v (Nh + m)
h + h
M =
h
=
+M
,
+ MR0
ih =
R0 1
,
+ MR0
iv =
40
(R0 1)
R0 ( + M)
N1 =
hbNh 1
number of secondary infections pro(Nh + m) v
duced by a single infectious mosquito during its lifespan.
N2 =
R0 = N1N2 = R0
41
STABILITY ANALYSIS
Linearizing around the disease-free equilibrium E1
DF (E1 ) = 0
0
0
(h + h )
v bNh
Nh +m
v
NhbA/
+m
h
hbA/v
Nh +m
Eigenvalues of DF (E1 ):
h
(h + h + v )
p
(h + h + v )2 4v (h + h )(1 R0 )
2
42
bhA/v
(Nh + m)v
iv + ih
Orbital derivative
bhA/v
V =
(1sh)iv (h +h)[1R0 (1iv )]ih 0
Nh + m
V = 0
(1 sh)iv = 0,
(1 sh)iv = 0,
ih = 0 if R0 < 1
iv ih = 0 if R0 = 1.
43
+ MR0
( + M)
+ hM + v R0
+M
+ MR0
B = 2h M
+ MR0
hM(R0 1)
+ v hr0 + v
+M
+ MR0
C = v 2hM(R0 1).
Routh-Hurwitz criterion (Gantmacher 1959), : all eigenvalues of p() have negative real parts if and only if
A > 0, B > 0, C > 0, and AB > C.
R0 > 1 B > 0, C > 0.
AB > C easy to verify.
E2 is locally asymptotically stable for R0 > 1.
44
45
(12)
fi(
x)
0, i 6= j.
xj
x)
fi(
0, i 6= j.
xj
n-dimensional cooperative and competitive systems behave like a dynamical flow in a (n-1)-dimensional space.
For n = 3
The solutions of a cooperative or competitive system
in a closed convex set D R3 that contains no equilibria are closed or approach a closed orbit when t 0.
47
fi
(
x) 0,
xj
i 6= j.
fi
(
x) 0,
xj
i 6= j.
The flow of a cooperative (competitive) system preserves for t > 0 (t < 0) the partial order m generated
x Rn | ixi 0, 1 i n}
by Km = {
x m y y x Km
48
1
H=0
0
0
1
0
0
0
1
K = {(sh, ih, iv ) R3 | sh 0, ih 0, iv 0 }
49
PERSISTENCE
(Butler & Waltman, 1986)
System (12) is persistent the solutions starting in int D remains at a positive distance from the
boundary of .
For system (10) the vector field points to the interior
of except in the sh axis. In this axis s0h = h(1sh)
sh(t) 1 as t E1 is the only equilibrium
in the boundary.
V = iv +
v (Nh + m)(1 + R0 )
ih
2bhA/v
50
51
COMPOUND MATRICES
(J.S. Muldowney, 1990)
A = n m matrix. Let ai1..ik,j1...jk determinant defined by the rows (i1, .., ik) and the columns (j1 , ..., jk),
1 i1 < i2 < < ik n, 1 j1 < j2 < <
jk m.
The kth multiplicative compound A(k) of A is the
Ckn Ckm matrix whose entries in lexicographic order are ai1..ik,j1...jk , where Ckn denotes the number of
combinations of n elements in groups of k elements.
For n k matrix with columns a1, a2, .., an
A(k) = a1 a2 ak .
If A is a n n matrix the kth additive compound
A[k] of A is the Ckn Ckn matrix
d
A[k] = (I + hA)(k) |h=0
dh
A[1] = A,
A[n]=Traza(A).
52
a23
a13
a11 + a22
,
a11 + a33
a12
A[2] = a32
a31
a22
a21 + a33
A/v
Nh
, av =
Nh + m
Nh + m
Lyapunov function:
i
i
(t)
(t)
h
h
Y (t),
Z(t)
V (X(t), Y (t), Z(t), sh (t), ih (t), iv (t)) = ((X(t),
iv (t)
iv (t)
||(X, Y, Z)|| = sup{|X|, |Y | + |Z|}
54
Along solutions:
V (t) = sup{|X(t),
ih(t)
(|Y (t)| + |Z(t)|)}
iv (t)
iv
h1 (t) = (h + ahb hiv + h + h ) + ahbhsh ,
ih
i0h i0v
h
h2 (t) = av bv (1 iv ) + h v av bv ih
iv
ih iv
i0v
iv i0h
h
ahbhsh = +h +h and
av bv (1iv ) = +v
ih ih
iv
iv
0
i
D+V (t) h + h V (t)
ih
V (t) V (0)ih (t)eht V (0)eh t 0 as t
(X(t), Y (t), Z(t)) 0 as t system (10)
has the property of stability of periodic orbits.
E2 is globally asymptotically stable for {(sh, 0, 0) :
0 sh 1}.
55
References
[1] E. Beretta and V. Capasso. On the general structure of epidemic systems. Global asymptotic stability. Comp & Maths. with Appls. 12A, 677-694
(1986).
[2] J.G. Butler and P. Waltman. Persistence in dynamical systems. Proc. Amer. Math. Soc. 96, 425-430
(1986).
[3] E. Coddington and N. Levinson. Theory of Ordinary Differential Equations, McGraw Hill, New
York, 1955.
[4] K.L. Cooke and J.A. Yorke. Some equations modelling growth processes and gonorrhea epidemics.
Math. Biosci. 16, 75-101 (1973).
[5] L. Esteva and C. Vargas. Analysis of a Dengue disease transmission model. Mathematical Biosciences
150, 131-151 (1998).
[6] F. R. Gantmacher. The Theory of Matrices,
Chelsea Publ. Co., New York, 1959.
[7] B.S. Goh. Global stability in a class of prey-predator
models. Bull. Math. Biol. 40, 525-533 (1978).
56
58