1 McKean-Vlasov SDE
{ d X t = β ( t , X t , μ t ) d t + σ ( t , X t , μ t ) W t X 0 = x 0 . ( 1.1 ) \newcommand{\dif}{\mathrm{d}} \left\{\begin{array}{l} {\dif X_{t}=\beta\left(t, X_{t}, \mu_{t}\right) \dif t+\sigma\left(t, X_{t}, \mu_{t}\right) W_{t}} \\ {X_{0}=x_{0}} \end{array} .(1.1)\right. { dXt=β(t,Xt,μt)dt+σ(t,Xt,μt)WtX0=x0.(1.1)
1.1 交互粒子系统逼近
对于系数中带分布的随机微分方程,分布的近似常用方法是粒子系统,用粒子系统逼近上述SDE,考虑N个满SDE(1.1)的粒子 X i , N X^{i,N} Xi,N:
{ d X t i , N = β ( θ , X t i , N , μ t i , N ) d t + σ ( θ , X t i , N , μ t i , N ) d W t i X 0 i , N = x 0 i , ( 1.2 ) \left\{\begin{array}{l} {d X_{t}^{i,N}=\beta\left(\theta, X_{t}^{i,N}, \mu_{t}^{i,N}\right) d t+\sigma\left(\theta, X_{t}^{i,N}, \mu_{t}^{i,N}\right) d W_{t}^i} \\ {X_{0}^{i,N}=x_{0}^{i}} \end{array},(1.2)\right. {
dXti,N=β(θ,Xti,N,μti,N)dt+σ(θ,Xti,N,μti,N)dWtiX0i,N=x0i,(1.2)
其中 μ t i , N : = 1 N ∑ i = 1 N δ X t i , N \mu_{t}^{i,N} := \frac{1}{N}\sum_{i=1}^{N}\delta_{ X_{t}^{i,N}}