Covid19：传染病模型SIR/SEIR on Network

github地址

SIR model on Network

We here reconsider the SIR model in network where $k$ is the number of neighbor and $v(t)$ is the average number of infected neighbor, $v$ is the average ratio of infected neighbor, $w(t)$ is the average number of neighbors recovered. We have:

$$\begin{gathered} \frac{d{s}_k}{dt}=-\beta kv{s}_k \\ \frac{d{i}_k}{dt}=\beta kv{s}_k-\gamma i_k \\ \frac{d{r}_k}{dt}=\gamma i_k \\ v(t)=\sum _{k=0}^{\infty }{q}_k{i}_k(t) \\ w(t)=\sum _{k=0}^{\infty }{q}_k{r}_k(t) \\ Takethederivativeofw(t)=\sum _{k=0}^{\infty }{q}_k{r}_k(t) \\ \frac{dw}{dt}=\sum _{k=0}^{\infty }{q}_k\frac{d{r}_k}{dt}=\gamma \sum _{k=0}^{\infty }{q}_k{i}_k=\gamma v(t) \\ \therefore v(t)=\frac{1}{\gamma }\frac{dw}{dt} \\ \frac{d{s}_k}{dt}=-\beta kv{s}_k\iff \frac{d{s}_k}{dt}=-\beta k\frac{1}{\gamma }\frac{dw}{dt}{s}_k \end{gathered}$$
We could solve these equation and get:

$$\begin{gathered} s_k=s_0{e}^{-\frac{\beta }{\gamma }kw} \\ =s_0[e^{-\frac{\beta }{\gamma }w}{]}^k \\ =s_0[u(t){]}^k \\ whereu(t)=e^{-\frac{\beta }{\gamma }w} \\ w(t)=-\frac{\gamma }{\beta }lnu \end{gathered}$$
With $s+i+r=1$, we could drive:

$$\begin{gathered} v(t)=\sum _{k=0}^{\infty }{q}_k{i}_k(t) \\ =\sum _{k=0}^{\infty }{q}_k(1-r_k-s_k) \\ =1-w(t)-s_0\sum _{k=0}^{\infty }{q}_k{u}^k \\ =1+\frac{\gamma }{\beta }lnu-s_0{G}_1(u) \\ \because w(t)=-\frac{\gamma }{\beta }lnu \\ \frac{dw}{dt}=\gamma v(t)\iff -\frac{\gamma }{\beta }\frac{1}{u}\frac{du}{dt}=-\gamma [1+\frac{\gamma }{\beta }lnu-s_0{G}_1(u)] \\ \therefore \frac{du}{dt}=-\beta u[1+\frac{\gamma }{\beta }lnu-s_0{G}_1(u)] \end{gathered}$$

Set $u=1-ϵ$ we could get

$$\begin{gathered} \frac{dϵ}{dt}=[\beta G_1^{\prime }(1)-\gamma ]ϵ \\ assume{s}_0=1 \\ u(t)=1-ϵ(t)=1-e^{(\beta G_1^{\prime }(1)-\gamma )t} \\ {s}_k(t)=u^k=[1-e^{(\beta G_1^{\prime }(1)-\gamma )t}{]}^k\simeq 1-e^{k(\beta G_1^{\prime }(1)-\gamma )t} \end{gathered}$$

$$\begin{gathered} Set{s}_0=1 \\ \frac{du}{dt}=-\beta u[1+\frac{\gamma }{\beta }lnu-G_1(u)] \\ s+i+r=1 \\ s=u^k \end{gathered}$$

With the above equations, we could fit the lines of the rate of susceptible, infectious and recovered people and get the parameter estimation of $\gamma$ and $\beta$. ($\gamma$ in the MGF method is just parameter for ${\tau }_i$, since $p(\tau )d\tau =\gamma e^{\gamma \tau }d\tau$, $\beta$ is the $T_{ij}$ which is simplified for the paired transmission)

If we set $s_k(t)=0$, we could get
$$\frac{\beta }{\gamma }=\frac{1}{G_1^{\prime }(1)}$$
Thus, average outbreak size here is $\frac{\beta }{\gamma }$.

Here, we could rewrite the model of SIR on network:
$$\begin{gathered} s_k(t)=u(t{)}^k \\ \frac{du}{dt}=-\beta u[1+\frac{\gamma }{\beta }lnu-s_0{G}_1(u)] \\ u(t)=1-e^{(\beta G_1^{\prime }(1)-\gamma )t} \\ \frac{d{i}_k}{dt}=ku(t{)}^{k-1}\frac{du}{dt}-\gamma i_k \\ \frac{d{r}_k}{dt}=\gamma i_k \\ s+i+r=1 \end{gathered}$$
Also, we could drive their iteration form with $S=sN,I=iN,R=rN$ and let $c$ be the number of people one contact:
$$\begin{gathered} U_n=U_{n-1}-\beta U_{n-1}[1+\frac{\gamma }{\beta }ln{U}_{n-1}-s_0{G}_1(U_{n-1})] \\ {S}_n=U_n^k \\ {I}_n=I_{n-1}-k{U}_{n-1}(t{)}^k\beta U_{n-1}[1+\frac{\gamma }{\beta }ln{U}_{n-1}-s_0{G}_1(U_{n-1})]-\gamma I_{n-1} \\ {R}_n=R_{n-1}+\gamma I_{n-1} \end{gathered}$$

SEIR model on Network

We here reconsider the SEIR model in network where $k$ is the number of neighbor and $v(t)$ is the average number of infected neighbor, $v$ is the average ratio of infected neighbor, $m(t)$ is the average number of exposed neighbor, $w(t)$ is the average number of neighbors recovered. We have:

$$\begin{gathered} \frac{d{s}_k}{dt}=-\beta kv{s}_k \\ \frac{d{e}_k}{dt}=\beta kv{s}_k-\alpha e_k \\ \frac{d{i}_k}{dt}=\alpha e_k-\gamma i_k \\ \frac{d{r}_k}{dt}=\gamma i_k \\ v(t)=\sum _{k=0}^{\infty }{q}_k{i}_k(t) \\ m(t)=\sum _{k=0}^{\infty }{q}_k{e}_k(t) \\ w(t)=\sum _{k=0}^{\infty }{q}_k{r}_k(t) \\ Takethederivativeofw(t)=\sum _{k=0}^{\infty }{q}_k{r}_k(t) \\ \frac{dw}{dt}=\sum _{k=0}^{\infty }{q}_k\frac{d{r}_k}{dt}=\gamma \sum _{k=0}^{\infty }{q}_k{i}_k=\gamma v(t) \\ \therefore v(t)=\frac{1}{\gamma }\frac{dw}{dt} \\ \frac{d{s}_k}{dt}=-\beta kv{s}_k\iff \frac{d{s}_k}{dt}=-\beta k\frac{1}{\gamma }\frac{dw}{dt}{s}_k \\ Takethederivativeofm(t)=\sum _{k=0}^{\infty }{q}_k{e}_k(t) \\ \frac{dm(t)}{dt}=\sum _{k=0}^{\infty }{q}_k\frac{d{e}_k}{dt}=\beta kv\sum _{k=0}^{\infty }{q}_k{s}_k-\sum _{k=0}^{\infty }{q}_k\alpha e_k \\ =\beta kv\sum _{k=0}^{\infty }{q}_k{s}_k-\alpha m(t) \\ Takethederivativeofv(t)=\sum _{k=0}^{\infty }{q}_k{i}_k(t) \\ \frac{dv(t)}{dt}=\sum _{k=0}^{\infty }{q}_k\frac{d{i}_k}{dt}=\alpha m(t)-\gamma v(t) \\ m(t)=\frac{1}{\alpha }[\frac{dv(t)}{dt}+\gamma v(t)] \end{gathered}$$

We could solve these equation and get:

$$\begin{gathered} s_k=s_0{e}^{-\frac{\beta }{\gamma }kw} \\ =s_0[e^{-\frac{\beta }{\gamma }w}{]}^k \\ =s_0[u(t){]}^k \\ whereu(t)=e^{-\frac{\beta }{\gamma }w} \\ w(t)=-\frac{\gamma }{\beta }lnu \end{gathered}$$

With $s+e+i+r=1$, we could drive:

$$\begin{gathered} v(t)=\sum _{k=0}^{\infty }{q}_k{i}_k(t) \\ =\sum _{k=0}^{\infty }{q}_k(1-r_k-e_k-s_k) \\ =1-w(t)-m(t)-s_0\sum _{k=0}^{\infty }{q}_k{u}^k \\ =1+\frac{\gamma }{\beta }lnu-m(t)-s_0{G}_1(u) \\ \therefore m(t)=1+\frac{\gamma }{\beta }lnu-s_0{G}_1(u)-v(t) \\ \because m(t)=\frac{1}{\alpha }[\frac{dv(t)}{dt}+\gamma v(t)] \\ \therefore (\frac{\gamma }{\alpha }+1)v(t)+\frac{1}{\alpha }\frac{dv(t)}{dt}=1+\frac{\gamma }{\beta }lnu-s_0{G}_1(u) \\ ForODEfunction\frac{dv}{dt}+p(t)v=q(t) \\ Thesolutionisv(t)=e^{-\int p(x)dx}C+e^{-\int p(x)dx}\int q(x)e^{\int p(x)dx}dx \\ Herep(t)=\gamma +\alpha \\ q(t)=[1+\frac{\gamma }{\beta }lnu-s_0{G}_1(u)]\alpha \\ \because w(t)=-\frac{\gamma }{\beta }lnu \\ \frac{dw}{dt}=\gamma v(t)\iff -\frac{\gamma }{\beta }\frac{1}{u}\frac{du}{dt}=-\gamma v(t) \\ \therefore \frac{du}{dt}=\beta uv(t) \\ =\beta u{e}^{-\int (\gamma +\alpha )dx}C+e^{-\int (\gamma +\alpha )dx}\int [[1+\frac{\gamma }{\beta }lnu-s_0{G}_1(u)]\alpha ]e^{\int (\gamma +\alpha )dx}dx \\ =C\beta u{e}^{-(\gamma +\alpha )t}+e^{-(\gamma +\alpha )t}\int [[1+\frac{\gamma }{\beta }lnu(t)-s_0{G}_1(u(t))]\alpha ]e^{(\gamma +\alpha )t}dt \end{gathered}$$

Assume$s_0=1$:

$$\begin{gathered} \frac{du}{dt}=C\beta u{e}^{-(\gamma +\alpha )t}+e^{-(\gamma +\alpha )t}\int [[1+\frac{\gamma }{\beta }lnu(t)-s_0{G}_1(u(t))]\alpha ]e^{(\gamma +\alpha )t}dt \\ {s}_k(t)=u^k \\ s+e+i+r=1 \end{gathered}$$

With the above equations, we could fit the lines of the rate of susceptible, infectious and recovered people and get the parameter estimation of $\gamma$ ,$\beta$, and $\alpha$. ($\gamma$ in the MGF method is just parameter for ${\tau }_i$, since $p(\tau )d\tau =\gamma e^{\gamma \tau }d\tau$, $\beta$ is the $T_{ij}$ which is simplified for the paired transmission, $\alpha$ is the rate that exposed people become sick)

Here, we could rewrite the model of SEIR on network:
$$\begin{gathered} s_k(t)=u(t{)}^k \\ \frac{du}{dt}=C\beta u{e}^{-(\gamma +\alpha )t}+e^{-(\gamma +\alpha )t}\int [[1+\frac{\gamma }{\beta }lnu(t)-s_0{G}_1(u(t))]\alpha ]e^{(\gamma +\alpha )t}dt \\ \frac{d{e}_k}{dt}=ku(t{)}^{k-1}\frac{du}{dt}-\alpha e_k \\ \frac{d{i}_k}{dt}=\alpha e_k-\gamma i_k \\ \frac{d{r}_k}{dt}=\gamma i_k \\ s+e+i+r=1 \end{gathered}$$
Also, we could drive their iteration form with $S=sN,E=eN,I=iN,R=rN$ and let $c$ be the number of people one contact:
$$\begin{gathered} U_n-\frac{\alpha \gamma }{\beta }ln{U}_n+\alpha G_1(U_n)=U_{n-1}-\frac{\alpha \gamma }{\beta }ln{U}_{n-1}+\alpha G_1(U_{n-1})+C\beta U_{n-1}{e}^{-(\gamma +\alpha )} \\ {S}_n=U_n^k \\ {E}_n=E_{n-1}+k{U}_{n-1}(t{)}^{k-1}[C\beta U_{n-1}{e}^{-(\gamma +\alpha )}+[\frac{\gamma }{\beta }ln{U}_{n-1}-\frac{\gamma }{\beta }ln{U}_{n-2}-s_0\alpha [G_1(U_{n-1})-G_1(U_{n-2})]-\alpha E_{n-1} \\ {I}_n=I_{n-1}+\alpha E_{n-1}-\gamma I_{n-1} \\ {R}_n=R_{n-1}+\gamma I_{n-1} \end{gathered}$$

.
欢迎关注二幺子的知识输出通道：