EPIDEMICS OVER COMPLEX NETWORKS: ANALYSIS OF EXACT AND APPROXIMATE MODELS
2.3 Results on the Nonlinear MFA Model
The nonlinear mean-field approximation has been extensively studied in the literature for different propagation models. We review the most important results here, starting from the SIS epidemics.
2.3.1 SIS
It is straightforward to see that the linear model upper bounds the nonlinear one, as follows.
๐๐(๐ก+1)=(1โ๐ฟ)๐๐(๐ก) + (1โ (1โ๐ฟ)๐๐(๐ก)) 1โ ร
๐โ๐๐
(1โ๐ฝ ๐๐(๐ก))
!
โค(1โ๐ฟ)๐๐(๐ก) + 1โ ร
๐โ๐๐
(1โ ๐ฝ ๐๐(๐ก))
!
โค(1โ๐ฟ)๐๐(๐ก) +๐ฝ ร
๐โ๐๐
๐๐(๐ก)
!
For two real-valued column vectors๐ข, ๐ฃ โR๐, we use the notation๐ข ๐ฃto indicate ๐ข๐ โค ๐ฃ๐ for all ๐ โ {1, . . . , ๐}, and ๐ข โบ ๐ฃ, if the inequalities are strict. Defining ๐(๐ก) = [๐1(๐ก), . . . , ๐๐(๐ก)]๐, we have
๐(๐ก+1) ( (1โ๐ฟ)๐ผ๐+ ๐ฝ ๐ด)๐(๐ก), (2.26) which leads to the following well-known result.
Proposition 1. If ๐ฝ๐max๐ฟ(๐ด) < 1, the origin is a globally asymptotically stable fixed point for both the linear SIS model(2.8)and the nonlinear SIS model(2.5).
The origin, the trivial fixed point of the nonlinear model, is unstable when๐๐ ๐๐ฅ( (1โ ๐ฟ)๐ผ๐+๐ฝ ๐ด) > 1. Moreover, if so, it is not clear in general whether there exists any
Figure 2.3: Summary of known results for different models. The results have been illustrated as a function of ๐ฝ๐๐๐ ๐ฅ๐ฟ (๐ด). MC stands for the Markov chain model. MFA stands for the mean-field approximation (the nonlinear model).
other fixed point, or how many fixed points there are. It has been shown in the literature (e.g., in [5]) that there exists a unique nontrivial fixed point, and it is stable.
Theorem 2. If ๐ฝ๐max๐ฟ(๐ด) > 1, the nonlinear SIS model(2.5)has a second unique fixed point. Furthermore, the fixed point is globally asymptotically stable from all initial points (except the origin).
2.3.2 SIRS/SEIRS/Immune-Admitting-SIS
Similar to the previous (immune-free) SIS model, for the immune-admitting-SIS, SIRS, and SEIRS epidemics, the linear model is an upper-bound on the nonlinear one, and therefore the origin is stable for the nonlinear model when the linear model is stable.
Proposition 3. If ๐ฝ๐max๐ฟ(๐ด) < 1, the origin is a globally asymptotically stable fixed point for both the linear model and the nonlinear model for immune-admitting-SIS, SIRS, and SEIRS epidemics.
In this case, when the origin is not stable, even though there still exists a unique nontrivial fixed point, it is not stable in general [5, 16].
Theorem 4. If ๐ฝ๐max๐ฟ(๐ด)
> 1, the nonlinear model for immune-admitting-SIS, SIRS, and SEIRS epidemics has a second unique fixed point.
The following is an example of an unstable nontrivial fixed point for the immune- admitting-SIS model [4, p. 64].
A=ยฉ
ยญ
ยญ
ยซ
0 1 1 1 0 0 1 0 0
ยช
ยฎ
ยฎ
ยฎ
ยฌ
๐ฝ=0.9 ๐ฟ =0.9 (2.27)
The nontrivial fixed point of the system above is๐โ =(0.286,0.222,0.222)๐. The linearized model around๐โ is
ยฉ
ยญ
ยญ
ยซ
โ0.260 0.514 0.514 0.700 โ0.157 0 0.700 0 โ0.157
ยช
ยฎ
ยฎ
ยฎ
ยฌ ,
which has an eigenvalue ofโ1.059, which means that๐โ is not locally stable. It can be shown that for any initial condition other the origin and๐โ,๐(๐ก)converges to a cycle.
Even though the nontrivial fixed point is not stable in general, it is known to be stable with high probability for a general family of random graphs [4, p. 66].
Theorem 5 (Ahn and Hassibi [5]). Suppose that ๐บ(๐) is a random graph with ๐ vertices, and let๐(
๐)
min and๐(
๐)
maxdenote the minimum and maximum degree of๐บ(๐). If, for any fixed๐ >0,P( (๐(
๐)
min)2 > ๐ยท๐(
๐)
max)goes to1as๐goes to infinity, then, with high probability, the origin is unstable, and the second fixed point is locally stable, for any fixed ๐ฝand๐ฟ.
One can think of several random graph models that satisfy the condition of Theorem 5. For example, if the random graph has a uniform degree that grows with๐, then the minimum degree and maximum degree are identical and the ratio
๐2
min
๐max
=๐will grow with any๐and exceed๐with high probability. Similarly, for random graphs where the degree distribution of the nodes are identical and concentrate, so that we can expect the maximum degree and the minimum degree to be proportional to the expected degree,
๐2
min
๐max grows if the expected degree increases unbounded with๐. In particular, the Erdรถs-Rรฉnyi random graph๐บ(๐) =๐บ(๐, ๐(๐))has identical degree distribution, and we have the following result [4, p. 69].
Corollary 6(Ahn and Hassibi [5]). Consider an Erdรถs-Rรฉnyi random graph๐บ(๐) = ๐บ(๐, ๐(๐))with ๐(๐)=๐
log๐
๐ where๐ > 1is a constant. The nonlinear model is
locally unstable at the origin and has a locally stable nontrivial fixed point with high probability for any fixed๐ฝand๐ฟ.
Since ๐ = ๐log๐ ๐
for ๐ = 1 is also the threshold for connectivity, we can say that connected Erdรถs-Rรฉnyi graphs have a nontrivial stable fixed point with high probability.
The random geometric graph๐บ(๐) =๐บ(๐, ๐(๐))also has identical degree distribution if each node is distributed uniformly. Such random graphs have maximum and minimum degrees which are proportional to the expected degree with high probability if๐(๐)is smaller than the threshold of connectivity [167], and, similar to Erdรถs-Rรฉnyi graphs, it has high probability of having a nontrivial stable fixed point if the degree grows with๐.
2.3.3 SIV/SEIV (Infection-Dominant)
Since the linear model is always the Jacobian of the nonlinear one, its stability determines the local stability of the nonlinear model. However, global stability is harder to show. The following result summarizes the stability of the nonlinear model.
Proposition 7. The disease-free fixed point of the nonlinear model for the infection- dominant SIV and the infection-dominant SEIV epidemics is
a) locally stable, if ๐พ+๐๐พ ๐ฝ๐ฟ
๐๐ ๐๐ฅ(๐ด) <1, and b) globally stable, if ๐ฝ๐ฟ๐๐ ๐๐ฅ(๐ด) < 1.
When ๐พ๐พ+๐๐ฝ๐max๐ฟ(๐ด) >1, the main fixed point of the nonlinear map is not stable, and again, there exists a unique non-trivial fixed point. This result has been proven in [16] for the SIV case, and it extends to the SEIV model in a similar fashion.
Theorem 8. If ๐พ๐พ+๐
๐ฝ๐max(๐ด)
๐ฟ > 1, the nonlinear model for the infection-dominant SIV and the infection-dominant SEIV epidemics has a second unique fixed point.
The middle panel in Figure 2.3 shows a summary of the above results.
2.3.4 SIV/SEIV (Vaccination-Dominant)
It is natural to expect the vaccination-dominant models to be more stable than the infection-dominant ones. It turns out that these models are indeed more stable by a factor of 1/(1โ๐).
Proposition 9. The disease-free fixed point of the nonlinear model for the vaccination- dominant SIV and the vaccination-dominant SEIV epidemics is
a) locally stable, if(1โ๐)๐พ+๐๐พ ๐ฝ
๐ฟ๐๐ ๐๐ฅ(๐ด) < 1, and b) globally stable, if(1โ๐)๐ฝ
๐ฟ๐๐ ๐๐ฅ(๐ด) < 1. Theorem 10. If (1โ๐) ๐พ
๐พ+๐ ๐ฝ
๐ฟ๐๐ ๐๐ฅ(๐ด) > 1, the nonlinear model for the vaccination- dominant SIV and the vaccination-dominant SEIV epidemics has a second unique fixed point.
Both of the above results have been proven in [16] for the SIV case, and the proof extends to the SEIV case with some modification. The lower panel in Figure 2.3 shows a summary of these results.