github地址

# Disease Spreading on Network

We consider the number of connection in a network and build up a model to describe the epidemic disease spreading on network. First, we could consider a network with n nodes and random edges and k be the degree of a node which means the number of connections each person has with others. The distribution of the network could be described a degree distribution p k p_k , and G 0 ( t ) G_0(t) is the moment generating function of k.

According to the definition of MGF, G 0 ( t ) = E ( e t k ) = ∑ k = 0 ∞ p k e t k = ∑ k = 0 ∞ p k x k ( ∵ x = e t ) G_0(t)=E(e^{tk})=\sum_{k=0}^\infty p_k e^{tk}=\sum_{k=0}^\infty p_kx^k(\because x=e^t)

According to the normalization property, G ( 1 ) = ∑ k = 0 ∞ p k = 1 G(1)=\sum_{k=0}^\infty p_k=1

We could find p k p_k by getting the function of G 0 ( t ) G_0(t) , since p k = d k d X k G 0 ( x ) k ! ∣ x = 0 p_k=\frac{d^k}{dX_k}\frac{G_0(x)}{k!}|_{x=0}

We could the r-th moments of p k p_k by z r = E [ k r ] = ∑ k = 0 ∞ k r p k = [ ( x d d x ) r G 0 ( x ) ] ∣ x = 1 z_r=E[k^r]=\sum_{k=0}^\infty k^r p_k=[(x\frac{d}{dx})^rG_0(x)]|_{x=1} , for example, the first moment z = G 0 ′ ( 1 ) z=G_0'(1) .

For Y = ∑ i = 1 m k i ,     k i ∼ p − k Y=\sum_{i=1}^mk_i, ~~~k_i\sim p-k , the MGF of sum of degree of m independent nodes is G Y ( x ) = [ G 0 ( x ) ] m G_Y(x)=[G_0(x)]^m

If node c has k c k_c edges, it is k c k_c times as likely connect to A as B is, thus p k ′ ∝ k p k p_k' \propto k p_k , ∑ p k ′ = c ∑ k p k = 1 \sum p_k'=c\sum kp_k=1 , so p k ′ = k p k ∑ j p j = k p k z p_k'=\frac{kp_k}{\sum_{j}p_j}=\frac{kp_k}{z} . The MGF of the original and first neighbors is G 1 ′ ( x ) = ∑ p k ′ x k = ∑ k p k ∑ j p j x k = x G 0 ′ ( x ) G 0 ′ ( 1 ) G_1'(x)=\sum p_k'x^k=\sum \frac{kp_k}{\sum_{j}p_j}x^k=x\frac{G_0'(x)}{G_0'(1)} .
F o r   e x c e s s   d e g r e e , q k = ( k + 1 ) p k + 1 ∑ j p j = ( k + 1 ) p k + 1 z G 1 ( x ) = ∑ k = 0 ∞ q k x k = ∑ k = 0 ∞ ( k + 1 ) p k + 1 x k z = ∑ k ′ = 0 ∞ k ′ p k ′ x k − 1 z = G 0 ′ ( x ) G 0 ′ ( 1 ) For ~excess~ degree, q_k=\frac{(k+1)p_{k+1}}{\sum _j p_j}\\ =\frac{(k+1)p_{k+1}}{z}\\ G_1(x)=\sum_{k=0}^\infty q_kx^k=\sum_{k=0}^\infty\frac{(k+1)p_{k+1}x^k}{z}\\ =\sum_{k'=0}^\infty\frac{k'p_{k'}x^{k-1}}{z}\\ =\frac{G_0'(x)}{G_0'(1)}\\
The MGF for the second neighbor is G ( 2 ) = ∑ p k G ( 2 ) ( x ∣ k ) = ∑ p k [ G 1 ( x ) ] k = G 0 ( G 1 ( x ) ) G^{(2)}=\sum p_kG^{(2)}(x|k)=\sum p_k [G_1(x)]^k=G_0(G_1(x)) . Similarly, the number of m-th nearest neighbor is G ( m ) ( x ) = G ( m − 1 ) ( G 1 ( x ) )   f o r   m > 2 G^{(m)}(x)=G^{(m-1)}(G_1(x))~for~ m>2 . Thus, the average number of n-th neighbor is z n = G ( m ) ′ ( x ) ∣ x = 1 = G 0 ′ ( 1 ) ∗ ( G 1 ′ ( 1 ) ) ( m − 1 ) z_n=G^{(m)'}(x)|_{x=1}=G_0'(1)*(G_1^{'}(1))^{(m-1)} .

## Considering Transmissibility for Outbreak Analysis

Let r i j r_{ij} be the rate of disease-causing contacts between i and j. τ i \tau_i be the time duration that i remains infective. T i j T_{ij} be the probability of transmission which is unknown but should be a function of r i j r_{ij} and τ i \tau_i .

Assume r i j ∼ P r ( r ) , τ i ∼ P τ ( τ ) r_{ij} \sim P_r(r),\tau_i\sim P_{\tau}(\tau) and they are independent from each other. Within ξ t \xi_t time interval, i does not inject j with probability 1 − r i j ξ t 1-r_{ij}\xi_t . so in all T i ξ t \frac{T_i}{\xi_t} time interval, the probability of Not getting the disease is ( 1 − r i j ξ t ) T i ξ t (1-r_{ij}\xi_t)\frac{T_i}{\xi_t} .
P [ i   d o e s   n o t   i n f e c t   j ] = 1 − T i j = l i m ξ t → o ( 1 − r i j ξ t ) T i ξ t = e − r i j τ i ∴ T i j = 1 − e − r i j τ i T h e   a v e r a g e   t r a n s i m i s s i b i l i t y   i s T = E [ T i j ] = 1 − E [ e − r i j τ i ] = 1 − ∫ 0 ∞ ∫ 0 ∞ e − r t P r ( r ) P τ ( τ ) d r d τ P[i~does~not~infect ~j]=1-T_{ij}\\ =lim_{\xi_t\rightarrow o}(1-r_{ij}\xi_t)\frac{T_i}{\xi_t}\\ =e^{-r_{ij}\tau_i}\\ \therefore T_{ij}=1-e^{-r_{ij}\tau_i}\\ The~average~transimissibility~is T=E[T_{ij}]\\ =1-E[e^{-r_{ij}\tau_i}]\\ =1-\int_0^\infty \int_0^\infty e^{-rt}P_r(r)P_{\tau}(\tau)drd\tau\\
Then we consider the MGF as a function of transmissibility in which a node with k links, each is accessibly transmitted(occupied) with probability T T .
P m 0 = P [ m   o f   k   l i n k s   o c c u p i e d ] = C m k T m ( 1 − T ) k − m G 0 ( x ; T ) = E [ x m ] = E [ E [ x m ∣ k ] ] = ∑ k = 0 ∞ E [ x m ∣ k ] p k = ∑ k = 0 ∞ ∑ m = 0 k x m P m k P k = ∑ k = 0 ∞ ∑ m = 0 k C m k T m ( 1 − T ) k − m x m P k = ∑ k = 0 ∞ ∑ m = 0 k C m k ( x T ) m ( 1 − T ) k − m P k = ∑ k = 0 ∞ ( x T + 1 − T ) k P k = G 0 ( 1 + ( x − 1 ) T ) A n d   G 0 ( x ; 1 ) = G 0 ( x )   G 0 ( 1 ; T ) = G 0 ( 1 )   G 0 ′ ( 1 ; T ) = T G 0 ′ ( 1 ) S i m i l a r l y , G 1 ( x ; T ) = G 1 ( 1 + ( x − 1 ) T )   G 1 ( x ; 1 ) = G 1 ( x )   G 1 ( 1 ; T ) = G 1 ( 1 )   G 1 ′ ( 1 ; T ) = T G 1 ′ ( 1 ) P_m^0=P[m~of~k~links~occupied]=C^k_mT^m(1-T)^{k-m}\\ G_0(x;T)=E[x^m]=E[E[x^m|k]]=\sum_{k=0}^\infty E[x^m|k]p_k\\ =\sum_{k=0}^\infty\sum_{m=0}^k x^m P_m^kP_k\\ =\sum_{k=0}^\infty\sum_{m=0}^kC^k_mT^m(1-T)^{k-m}x^mP_k\\ =\sum_{k=0}^\infty\sum_{m=0}^kC^k_m(xT)^m(1-T)^{k-m}P_k\\ =\sum_{k=0}^\infty(xT+1-T)^kP_k\\ =G_0(1+(x-1)T)\\ And~G_0(x;1)=G_0(x)\\ ~G_0(1;T)=G_0(1)\\ ~G_0'(1;T)=TG_0'(1)\\ Similarly, G_1(x;T)=G_1(1+(x-1)T)\\ ~G_1(x;1)=G_1(x)\\ ~G_1(1;T)=G_1(1)\\ ~G_1'(1;T)=TG_1'(1)\\
Let P s ( T ) P_s(T) be the distribution of outbreak size s s , H 0 ( x ; T ) H_0(x;T) is the corresponding MGF. By definition, H 1 ( x ; T ) H_1(x;T) is the MGF for the outbreak size reached by following a random edge.
KaTeX parse error: No such environment: align at position 8: \begin{̲a̲l̲i̲g̲n̲}̲ H_0(x;T)&=\sum…

## Scale-free Network

A scale-free network is a network whose degree distribution follows a power law distribution. This type of network has three important features: 1) has super nodes which have many more connections than others and this is called “preferential attachment”; 2) the number of links connecting to a node follows a power-law distribution or clustering coefficient distribution; 3) the degree distribution has a heavy tail. Since many research find that social network is a specific kind of scale-free network, we here study the spread of covid-19 in a scale-free network and conduct a experiment based on Shenzhen data later.

For the degree distribution to be power-law distribution, p k = C k − ν p_k=Ck^{-\nu} . For normalization property, G ( 1 ) = ∑ k = 0 ∞ C k − ν = 1 G(1)=\sum_{k=0}^\infty Ck^{-\nu}=1 so C = 1 ∑ k = 0 ∞ k − ν C=\frac{1} {\sum_{k=0}^ \infty k^{-\nu}} .
G 0 ( x ) = ∑ k = 0 ∞ C k − ν x k = ∑ k = 0 ∞ k − ν x k ∑ k = 0 ∞ k − ν z 1 ( x ) = G 0 ′ ( 1 ) = ∑ k = 0 ∞ k ∗ k − ν x k − 1 ∑ k = 0 ∞ k − ν ∣ x = 1 = ∑ k = 0 ∞ k − ν + 1 ∑ k = 0 ∞ k − ν = ξ ( ν − 1 ) ξ ( ν ) z 2 ( x ) = G 0 ′ ′ ( 1 ) = ∑ k = 0 ∞ k ∗ ( k − 1 ) ∗ k − ν x k − 2 ∑ k = 0 ∞ k − ν ∣ x = 1 = ∑ k = 0 ∞ k − ν + 2 − k − ν + 1 ∑ k = 0 ∞ k − ν = ξ ( ν − 2 ) − ξ ( ν − 1 ) ξ ( ν ) G_0(x)=\sum_{k=0}^\infty Ck^{-\nu}x^k=\frac{\sum_{k=0}^ \infty k^{-\nu}x^k}{\sum_{k=0}^ \infty k^{-\nu}}\\ z_1(x)=G_0'(1)=\frac{\sum_{k=0}^ \infty k*k^{-\nu}x^{k-1}}{\sum_{k=0}^ \infty k^{-\nu}}|_{x=1}\\ =\frac{\sum_{k=0}^ \infty k^{-\nu+1}}{\sum_{k=0}^ \infty k^{-\nu}}\\ =\frac{\xi(\nu-1)}{\xi(\nu)}\\ z_2(x)=G_0''(1)=\frac{\sum_{k=0}^ \infty k*(k-1)*k^{-\nu}x^{k-2}}{\sum_{k=0}^ \infty k^{-\nu}}|_{x=1}\\ =\frac{\sum_{k=0}^ \infty k^{-\nu+2}-k^{-\nu+1}}{\sum_{k=0}^ \infty k^{-\nu}}\\ =\frac{\xi(\nu-2)-\xi(\nu-1)}{\xi(\nu)}\\
Assume r i j ∼ P r ( r ) = 1 r m   0 ≤ r ≤ r m , τ i ∼ P τ ( τ ) = 2 τ τ m 2 r_{ij} \sim P_r(r)=\frac{1}{r_m} ~0\leq r\leq r_m,\tau_i\sim P_{\tau}(\tau)=\frac{2\tau}{\tau_m^2}
T h e   a v e r a g e   t r a n s i m i s s i b i l i t y   i s T = 1 − ∫ 0 ∞ ∫ 0 ∞ e − r t P r ( r ) P τ ( τ ) d r d τ = 1 − 2 r m 2 τ m 2 ( e − r m τ m − 1 + r m τ m ) I n   t h i s   c a s e , T c = G 0 ′ ( 1 ) G 0 ′ ′ ( 1 ) = ξ ( ν − 1 ) ξ ( ν − 2 ) − ξ ( ν − 1 ) A v e r a g e   o u t b r e a k   s i z e   E [ s ] = 1 + T ( ξ ( ν − 1 ) ) 2 ξ ( ν ) [ ( T + 1 ) ξ ( ν − 1 ) − T ξ ( ν − 2 ) ] The~average~transimissibility~is T =1-\int_0^\infty \int_0^\infty e^{-rt}P_r(r)P_{\tau}(\tau)drd\tau\\ =1-\frac{2}{r_m^2\tau_m^2}(e^{-r_m\tau_m}-1+r_m\tau_m)\\ In~this~case,T_c=\frac{G_0'(1)}{G_0''(1)}=\frac{\xi(\nu-1)}{\xi(\nu-2)-\xi(\nu-1)}\\ Average~outbreak~size~E[s]=1+\frac{T(\xi(\nu-1))^2}{\xi(\nu)[(T+1)\xi(\nu-1)-T\xi(\nu-2)]}\\
Here, we know the number of first neighbor and second neighbor, we would estimate the parameter in the power-law distribution by method of moment and maximizing log likelihood.
For sampling distribution of continuous power law
p k = ν − 1 k m i n ( k k m i n ) − ν w h e r e   c = ν − 1 k m i n − ν + 1 F o r   m − t h   m o m e n t   z m = ∫ x m i n ∞ x m ν − 1 x m i n ( x x m i n ) − ν = ν − 1 ν − 1 − m x m i n m F o r   m e t h o d   o f   m o m e n t s   E [ x ] = ν − 1 ν − 2 x m i n E [ x 2 ] = ν − 1 ν − 3 x m i n 2 v a r ( x ) = ν − 1 ( ν − 3 ) ( ν − 2 ) 2 x m i n 2 ∴ x ^ m i n = ∑ x i 2 ( ν − 3 ) ∑ x i ( ν − 2 ) ∴ ν ^   s a t i s f i e s ( ν − 2 ) 2 ( ν − 1 ) ( ν − 3 ) = n ∑ x i 2 ( ∑ x i ) 2 v ^ = − 2 ( ∑ x i ) 2 + 2 n ∑ x i 2 ± − n ( ∑ x i ) 2 ∑ x i 2 + n 2 ( ∑ x i 2 ) 2 − ( ∑ x i ) 2 + n ∑ x i 2 p_k=\frac{\nu-1}{k_{min}}(\frac{k}{k_{min}})^{-\nu}where ~c=\frac{\nu-1}{k_{min}^{-\nu+1}}\\ For~m-th~moment ~z_m=\int_{x_{min}}^\infty x^m\frac{\nu-1}{x_{min}}(\frac{x}{x_{min}})^{-\nu}=\frac{\nu-1}{\nu-1-m}x_{min}^m\\ For ~method~of~moments~E[x]=\frac{\nu-1}{\nu-2}x_{min}\\ E[x^2]=\frac{\nu-1}{\nu-3}x_{min}^2\\ var(x)=\frac{\nu-1}{(\nu-3)(\nu-2)^2}x_{min}^2\\ \therefore \hat{x}_{min}=\frac{\sum x_i^2(\nu-3)}{\sum x_i(\nu-2)}\\ \therefore \hat{\nu}~satisfies \frac{(\nu-2)^2}{(\nu-1)(\nu-3)}=n\frac{\sum x_i^2}{(\sum x_i)^2}\\ \hat{v}=\frac{-2(\sum x_i)^2+2n\sum x_i^2 \pm \sqrt{-n(\sum x_i)^2 \sum x_i^2+n^2 (\sum x_i^2)^2}}{-(\sum x_i)^2+n\sum x_i^2}

F o r   m a x i m i z i n g   l o g − l i k e l i h o o d   e s t i m a t o r W i t h   c o n t i n u s   a s s u m p t i o n p ( x ) = ∏ i = 1 n ν − 1 x m i n ( x x m i n ) − ν l o g p ( x ) = n l o g ( ν − 1 ) − n l o g x m i n − ν ∑ i = 1 ∞ l o g x i x m i n ∂ l o g p ( x ) ∂ x m i n = − n x m i n + ν x m i n ∑ i = 1 ∞ 1 x i = 0 ∂ l o g p ( x ) ∂ ν = n ν − 1 − ∑ i = 1 ∞ l o g x i x m i n ν = 1 + n [ ∑ i = 1 n l n x i x m i n ] − 1 W i t h   d i s c r e t e   a s s u m t i o n ν = 1 + n [ ∑ i = 1 n l n x i x m i n − 1 2 ] − 1 L e t   u s   a s s u m e   x m i n = 1 , ν ^ = 1 + n l n ( ∑ x i ) For~maximizing ~log-likelihood~estimator\\ With~continus~assumption\\ p(x)=\prod_{i=1}^n\frac{\nu-1}{x_{min}}(\frac{x}{x_{min}})^{-\nu}\\ logp(x)=nlog(\nu-1)-nlogx_{min}-\nu\sum_{i=1}^\infty log\frac{x_i}{x_{min}}\\ \frac{\partial logp(x)}{\partial x_{min}}=\frac{-n}{x_{min}}+\frac{\nu}{x_{min}}\sum_{i=1}^\infty \frac{1}{x_i}=0\\ \frac{\partial logp(x)}{\partial \nu}=\frac{n}{\nu-1}-\sum_{i=1}^\infty log\frac{x_i}{x_{min}}\\ \nu=1+n[\sum_{i=1}^nln\frac{x_i}{x_{min}}]^{-1}\\ With~discrete~assumtion\\ \nu=1+n[\sum_{i=1}^nln\frac{x_i}{x_{min}-\frac{1}{2}}]^{-1}\\ Let~us~assume~x_{min}=1,\hat{\nu}=1+\frac{n}{ln(\sum x_i)}\\

T c ^ = G 0 ′ ( 1 ) G 0 ′ ′ ( 1 ) = ξ ( ν − 1 ) ξ ( ν − 2 ) − ξ ( ν − 1 ) = ν ^ − 3 ν ^ − 2 1 x ^ m i n E [ s ] ^ = 1 − ( ν ^ − 1 ) ( ν ^ − 3 ) ν ^ − 2 \hat{T_c}=\frac{G_0'(1)}{G_0''(1)}=\frac{\xi(\nu-1)}{\xi(\nu-2)-\xi(\nu-1)} \\ =\frac{\hat{\nu}-3}{\hat{\nu}-2}\frac{1}{\hat{x}_{min}}\\ \hat{E[s]}=1-\frac{(\hat{\nu}-1)(\hat{\nu}-3)}{\hat{\nu}-2}\\

# Reference

.                                                                           ©️2020 CSDN 皮肤主题: 数字20 设计师:CSDN官方博客