我们引用Tass的文章:
- Tass, P. A., et al. “Long-lasting desynchronization in rat hippocampal slice induced by coordinated reset stimulation.” Physical Review E 80.1 (2009): 011902.
- Hauptmann, Christian, and Peter A. Tass. “Restoration of segregated, physiological neuronal connectivity by desynchronizing stimulation.” Journal of neural engineering7.5 (2010): 056008.
Tam, H. C., Emily SC Ching, and Pik-Yin Lai. “Reconstructing networks from dynamics with correlated noise.” Physica A: Statistical Mechanics and its Applications 502 (2018): 106-122.
The time evolution of the node variables is given by a set of coupled differential equations:
d
x
i
d
t
=
f
i
(
x
i
)
+
∑
j
≠
i
A
i
j
h
(
x
i
,
x
j
)
+
η
i
\frac{dx_i}{dt}=f_i(x_i)+\sum_{j\neq i}A_{ij}h(x_i,x_j)+\eta_i
dtdxi=fi(xi)+j̸=i∑Aijh(xi,xj)+ηi
External influences on the system are modelled by a Gaussian correlated noise
η
i
\eta_i
ηi that is generated by the Ornstein-Uhlenbeck process:
τ
n
d
η
i
d
t
=
−
η
i
+
ξ
i
\tau_n\frac{d\eta_i}{dt}=-\eta_i+\xi_i
τndtdηi=−ηi+ξi
where
ξ
i
(
t
)
\xi_i(t)
ξi(t) is a zero-mean Gaussian white noise:
<
ξ
i
(
t
)
>
=
0
,
<
ξ
i
(
t
)
ξ
j
(
t
’
)
>
=
2
D
i
j
δ
(
t
−
t
’
)
<\xi_i(t)>=0,<\xi_i(t)\xi_j(t’)>=2D_{ij}\delta(t-t’)
<ξi(t)>=0,<ξi(t)ξj(t’)>=2Dijδ(t−t’)
Thus we have
<
η
i
(
t
)
>
=
0
,
<
η
i
(
t
)
η
j
(
t
’
)
>
=
D
i
j
τ
n
e
−
∣
t
−
t
’
∣
/
τ
n
<\eta_i(t)>=0,<\eta_i(t)\eta_j(t’)>=\frac{D_{ij}}{\tau_n}e^{-|t-t’|/\tau_n}
<ηi(t)>=0,<ηi(t)ηj(t’)>=τnDije−∣t−t’∣/τn
线性化可得:
d
d
t
δ
x
=
Q
δ
x
+
η
i
\frac{d}{dt}\delta x=Q\delta x+\eta_i
dtdδx=Qδx+ηi
对上式积分可得:
δ
x
(
t
)
=
e
Q
t
δ
x
(
0
)
+
∫
0
t
e
(
t
−
t
’
)
Q
η
i
(
t
’
)
d
t
’
\delta x(t)=e^{Qt}\delta x(0)+\int_0^te^{(t-t’)Q}\eta_i(t’)dt’
δx(t)=eQtδx(0)+∫0te(t−t’)Qηi(t’)dt’
Hence we obtain
K
τ
=
<
δ
x
(
t
+
τ
)
δ
x
(
t
)
T
>
K_{\tau}=<\delta x(t+\tau)\delta x(t)^T>
Kτ=<δx(t+τ)δx(t)T>
Q K 0 + K 0 Q T + D ( I − τ n Q T ) − 1 + ( I − τ n Q ) − 1 D T = 0 QK_0+K_0Q^T+D(I-\tau_nQ^T)^{-1}+(I-\tau_nQ)^{-1}D^T=0 QK0+K0QT+D(I−τnQT)−1+(I−τnQ)−1DT=0
K τ = e τ Q K 0 + J ( τ ) K_{\tau}=e^{\tau Q}K_0+J(\tau) Kτ=eτQK0+J(τ)
J ( τ ) = ( e τ Q − e − τ / τ n I ) U + τ e τ Q V J(\tau)=(e^{\tau Q}-e^{-\tau/\tau_n}I)U+\tau e^{\tau Q}V J(τ)=(eτQ−e−τ/τnI)U+τeτQV
U = τ n R D ( I − τ n Q T ) − 1 U=\tau_nRD(I-\tau_nQ^T)^{-1} U=τnRD(I−τnQT)−1
V = [ I − ( I + τ n Q ) R ] R D ( I − τ n Q T ) − 1 V=[I-(I+\tau_nQ)R]RD(I-\tau_nQ^T)^{-1} V=[I−(I+τnQ)R]RD(I−τnQT)−1
R = ( I + τ n Q ) − 1 R=(I+\tau_nQ)^{-1} R=(I+τnQ)−1
Replacing
τ
\tau
τ by
2
τ
2\tau
2τ and then by
3
τ
3\tau
3τ and after some algebra, we obtain
K
2
τ
=
S
K
τ
−
T
K
0
K_{2\tau}=SK_{\tau}-TK_0
K2τ=SKτ−TK0
K 3 τ = S K 2 τ − T K τ K_{3\tau}=SK_{2\tau}-TK_{\tau} K3τ=SK2τ−TKτ
S = e Q τ + e − τ / τ n I S=e^{Q\tau}+e^{-\tau/\tau_n}I S=eQτ+e−τ/τnI
T = e − τ / τ n e Q τ T=e^{-\tau/\tau_n}e^{Q\tau} T=e−τ/τneQτ
Solving Eqs, we then obtain
T
=
(
K
3
τ
−
K
2
τ
K
τ
−
1
K
2
τ
)
(
K
0
K
τ
−
1
K
2
τ
−
K
τ
)
−
1
T=(K_{3\tau}-K_{2\tau}K_{\tau}^{-1}K_{2\tau})(K_{0}K_{\tau}^{-1}K_{2\tau}-K_{\tau})^{-1}
T=(K3τ−K2τKτ−1K2τ)(K0Kτ−1K2τ−Kτ)−1
S = ( K 2 τ + T K τ ) K 0 − 1 S=(K_{2\tau}+TK_{\tau})K_{0}^{-1} S=(K2τ+TKτ)K0−1
S = e τ / τ n T + e − τ / τ n I S=e^{\tau/\tau_n}T+e^{-\tau/\tau_n}I S=eτ/τnT+e−τ/τnI
to estimate the value of e τ / τ n e^{\tau/\tau_n} eτ/τn and thus τ n \tau_n τn by a least-square fit of the diagonal elements of S S S, denoted by μ \mu μ,
Q = 1 τ l o g ( μ T ) = 1 τ [ l o g ( μ ) + l o g T ] Q=\frac{1}{\tau}log(\mu T)=\frac{1}{\tau}[log(\mu)+logT] Q=τ1log(μT)=τ1[log(μ)+logT]