UA MATH565C 随机微分方程III Ito Isometry
UA MATH565C 隨機微分方程III Ito Isometry
定義ft,gtf_t,g_tft?,gt?是step process,?0=t0<t1?<tn=t\forall 0=t_0<t_1 \cdots < t_n = t?0=t0?<t1??<tn?=t,
ft=fj,t∈[tj,tj+1),j=0,?,n?1ft=0,t≥tnf_t = f_j, t \in [t_j,t_{j+1}),j=0,\cdots,n-1 \\ f_t = 0, t\ge t_nft?=fj?,t∈[tj?,tj+1?),j=0,?,n?1ft?=0,t≥tn?
gt=gj,t∈[tj,tj+1),j=0,?,n?1gt=0,t≥tng_t = g_j, t \in [t_j,t_{j+1}),j=0,\cdots,n-1 \\ g_t = 0, t\ge t_ngt?=gj?,t∈[tj?,tj+1?),j=0,?,n?1gt?=0,t≥tn?
則Ito Isometry指的是
E[∫0tfsdWs∫0tgsdWs]=∫0tE[fsgs]dsE \left[ \int_0^t f_s dW_s \int_0^t g_s dW_s \right] = \int_0^t E \left[ f_s g_s \right]dsE[∫0t?fs?dWs?∫0t?gs?dWs?]=∫0t?E[fs?gs?]ds
證明
E[∫0tfsdWs∫0tgsdWs]=E[∑j=0n?1fjΔjWt∑k=0n?1gkΔkWt]=∑j=0n?1∑k=0n?1E[fjgkΔkWtΔjWt]E \left[ \int_0^t f_s dW_s \int_0^t g_s dW_s \right] \\= E \left[ \sum_{j=0}^{n-1} f_j\Delta_j W_t \sum_{k=0}^{n-1} g_k\Delta_k W_t \right] \\ = \sum_{j=0}^{n-1} \sum_{k=0}^{n-1} E \left[ f_j g_k\Delta_k W_t \Delta_j W_t\right] E[∫0t?fs?dWs?∫0t?gs?dWs?]=E[j=0∑n?1?fj?Δj?Wt?k=0∑n?1?gk?Δk?Wt?]=j=0∑n?1?k=0∑n?1?E[fj?gk?Δk?Wt?Δj?Wt?]
如果j<kj<kj<k
E[fjgkΔkWtΔjWt]=E[fjgkΔjWt]E[ΔkWt]=0E \left[ f_j g_k\Delta_k W_t \Delta_j W_t\right] = E \left[ f_j g_k \Delta_j W_t\right] E[\Delta_k W_t ]=0E[fj?gk?Δk?Wt?Δj?Wt?]=E[fj?gk?Δj?Wt?]E[Δk?Wt?]=0
這里獨立性的用法與性質2一樣;如果j=kj=kj=k
E[fjgjΔjWtΔjWt]=E[fjgj]E[ΔjWtΔjWt]=E[fjgj]ΔjtE \left[ f_j g_j\Delta_j W_t \Delta_j W_t\right] = E \left[ f_j g_j\right] E[\Delta_j W_t \Delta_j W_t] = E \left[ f_j g_j\right] \Delta_j tE[fj?gj?Δj?Wt?Δj?Wt?]=E[fj?gj?]E[Δj?Wt?Δj?Wt?]=E[fj?gj?]Δj?t
所以
E[∫0tfsdWs∫0tgsdWs]=∑j=0n?1E[fjgj]ΔjtE \left[ \int_0^t f_s dW_s \int_0^t g_s dW_s \right] =\sum_{j=0}^{n-1} E \left[ f_j g_j\right] \Delta_j tE[∫0t?fs?dWs?∫0t?gs?dWs?]=j=0∑n?1?E[fj?gj?]Δj?t
顯然右邊這個是簡單可測函數的Lebesgue積分:
∫0tE[fsgs]ds\int_0^t E \left[ f_s g_s \right]ds∫0t?E[fs?gs?]ds
綜上
E[∫0tfsdWs∫0tgsdWs]=∫0tE[fsgs]dsE \left[ \int_0^t f_s dW_s \int_0^t g_s dW_s \right] = \int_0^t E \left[ f_s g_s \right]dsE[∫0t?fs?dWs?∫0t?gs?dWs?]=∫0t?E[fs?gs?]ds
證畢。
關于這個性質再做一些評注。左邊這個表達式E[∫0tfsdWs∫0tgsdWs]E \left[ \int_0^t f_s dW_s \int_0^t g_s dW_s \right]E[∫0t?fs?dWs?∫0t?gs?dWs?],展開來寫是
∫Ω[∫0tfsdWs∫0tgsdWs]dP\int_{\Omega} \left[ \int_0^t f_s dW_s \int_0^t g_s dW_s \right] dP∫Ω?[∫0t?fs?dWs?∫0t?gs?dWs?]dP
它可以看成是Hilbert空間L2(Ω,Ft,P)L^2(\Omega,\mathcal{F}_t,P)L2(Ω,Ft?,P)上的內積。右邊的表達式∫0tE[fsgs]ds\int_0^t E \left[ f_s g_s \right]ds∫0t?E[fs?gs?]ds展開來寫是
∫0t∫Ωfsgsdλ?P\int_0^t\int_{\Omega}f_sg_s d\lambda \otimes P∫0t?∫Ω?fs?gs?dλ?P
這個是Hilbert空間L2(Ω×[0,∞),F?B([0,∞)),λ?P)L^2(\Omega\times[0,\infty),\mathcal{F} \otimes \mathcal{B}([0,\infty)),\lambda \otimes P)L2(Ω×[0,∞),F?B([0,∞)),λ?P)上的積分。這個Hilbert空間的σ\sigmaσ代數其實就是是我們一開始定義的隨機過程的那個σ\sigmaσ代數。Isometry的含義是等距同構,假設Hilbert空間L2(Ω,Ft,P)L^2(\Omega,\mathcal{F}_t,P)L2(Ω,Ft?,P)就用那個內積導出的距離,然后用右邊的表達式展開的那個積分作為Hilbert空間L2(Ω×[0,∞),F?B([0,∞)),λ?P)L^2(\Omega\times[0,\infty),\mathcal{F} \otimes \mathcal{B}([0,\infty)),\lambda \otimes P)L2(Ω×[0,∞),F?B([0,∞)),λ?P)上距離,考慮映射:
J0:L2(Ω×[0,∞),F?B([0,∞)),λ?P)→L2(Ω,Ft,P)J_0:L^2(\Omega\times[0,\infty),\mathcal{F} \otimes \mathcal{B}([0,\infty)),\lambda \otimes P) \to L^2(\Omega,\mathcal{F}_t,P)J0?:L2(Ω×[0,∞),F?B([0,∞)),λ?P)→L2(Ω,Ft?,P)
根據性質3,在上面兩個距離(fsf_sfs?與gsg_sgs?的距離)的定義下,這兩個空間是等距同構的,J0J_0J0?是他們的同構映射,所以性質3叫做Ito Isometry。
這個性質有一個非常常用的推論:如果f=gf=gf=g
E[∫0tfsdWs]2=∫0tE[fs]2dsE \left[ \int_0^t f_s dW_s \right]^2 = \int_0^t E \left[ f_s \right]^2dsE[∫0t?fs?dWs?]2=∫0t?E[fs?]2ds
總結
以上是生活随笔為你收集整理的UA MATH565C 随机微分方程III Ito Isometry的全部內容,希望文章能夠幫你解決所遇到的問題。
- 上一篇: UA MATH564 概率论VI 数理统
- 下一篇: UA MATH636 信息论9 Reed