文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > Finite Time--Horizon Risk Sensitive Control and the Robust Limit under a Quadratic Growth A
侵权投诉

Finite Time--Horizon Risk Sensitive Control and the Robust Limit under a Quadratic Growth A

The finite time--horizon risk sensitive limit problem for continuous, nonlinear systems is considered. Previous results are extended to cover more typical examples. In particular, the cost may grow quadratically, and the diffusion coefficient may depend on

Lipschitz bounds dependent on"; b; R; m were su cient to obtain the above result. Below however, we will need bounds independent of these parameters. These bounds will require more technical arguments. We now present such arguments. The rst bound is actually quite straight{forward.

Lemma 3.2 There exists e< 1 such that for all> e, V";b;R;m (s; x) MT (1+ jxj2) 8 s 2 0; T]; 8 x 2<n; and further, for 1,{optimal w 2 Ws satisfy 0 kwk2 ( 0;T]) MT (1+ jxj2): L 2

(28)

assumptions, it is quite conservative. If one has additional structure such as contractivity of f, a much smaller value for e can be obtained 29], 30]. Proof.

Remark 3.3 A value for e is given in the proof. Since this e is general for all systems satisfying the Suppressing certain arguments for notational simplicity, one has

jyt";bj2 jxj2+

Z

t

s

";b (yr )T f b+ wr] dr+

r

"

Z

2 s

t

Z tX";b aii dr (yr )T dBr+"2

s

where (see 26] for example) the stochastic integral is a square{integrable martingale. Then, using (A1:iii); (A2:iii), ";b Ejyt j2

jxj2+ KE jxj2+

By Gronwall's inequality, there exists Ct< 1 such that Consequently,

Z t";b j+ jy";b j2) dr+ ME jy";b jjwr j dr+" M 2(t? s) (jyr r r 2 s s Z t Z t 3K+ M";b j2 dr+" M 2+ K (t? s)+ M 2 2 2 2 s Ejyr 2 2 s Ejwr j dr:

Z

t

";b Ejyt j2 Ct (1+ jxj2+ kwk2 ( 0;T]) ): L 2

0 Jg";b;R (s; x; u; w) CT (1+ jxj2)+ (e2? 2 )kwk2 ( 0;T]) L 0; e which depend on T but not b; R and" 1. This yields the rst assertion. The for proper choice of CT second then follows by comparison with w 0. 2 2

The proof of continuity estimates uniform over b; R; m and" 1 are considerably more technical than the above results. In the deterministic setting, an L2{bound on{optimal w is su cient to yield equicontinuity estimates. However, in the stochastic case, the type of L2{bound obtained in the previous lemma is not su cient. This is due to the fact that L2 bounds over the sample space as well as the state space do not imply bounds on the expectation of the exponentiation of the state space L2{norm of w, and this is exactly what appears when using Gronwall{type estimates. (Recall we are only assuming Lipschitz conditions on the dynamics - not stability). However, one expects that if a set of Brownian paths does not diverge much from the origin,

then the near-optimal disturbance paths over that portion of the sample{space should not have\too large" L2{norms. Measurability conditions make it di cult to re ne that idea into a proof in the continuous{time case, so we work rst with discrete{time approximations, and then use well{known limit results to obtain the continuity estimate. 10

上一页第10页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email