节点文献
算子半群的逼近及其在参数连续Markov链中的应用
Approximation of Semigroups of Linear Operators and Applications to Continuous-Time Markov Chains
【作者】 赵文强;
【导师】 李扬荣;
【作者基本信息】 西南师范大学 , 基础数学, 2004, 硕士
【摘要】 关于Markov过程理论的研究,历来有概率方法和分析方法。近年来,数学家们以算子半群理论作为工具来研究Markov过程理论,并取得了丰富的成果。本文着力使用分析的方法,以算子半群理论为工具,研究参数连续Markov链中的转移函数、q-矩阵、以及转移函数与它的q-矩阵之间的连续依赖关系。 我们知道转移函数是l1空间上的正的强连续压缩半群,但是转移函数一般来说不是l∞空间上的正的强连续压缩半群。因此Anderson[2]认为l∞空间太大了,不可能在其上得到一些有用的结果。而近年来发展起来的积分算子半群在参数连续Markov链中是否能得到应用?Li Y.R.在[33,36]中做了讨论。证明了转移函数是l∞空间上正的一次强压缩积分算子半群(我们称作Maxkov积分算子半群),得到了转移函数和l∞空间上正的一次强压缩积分算子半群之间的对应关系。一个自然的问题是:是否存在l∞空间上的一个充分大的子空间,使得转移函数是其上的正的强连续压缩半群?在Li Y.R.[33,36]的Markov积分算子半群理论的基础上,我们有如下结果: 定理4.2.2设P(t)=(pij(t))是q-矩阵Q的一q-函数,G(t)代表相对应的Markov积分算子半群,其生成元用Ω表示。定义 C1={x∈l∞;P(t)x是l∞空间上关于t≥0的连续函数}则转移函数P(t)=(pij(t))是C1上的正的强连续压缩半群,且其生成元Ω0是Ω在C1中的部分,也就是 D(Ω0)={x∈D(Ω);Ωx∈C1}并且对每一个x∈D(Ω0)有 Ω0x=Ωx. 定理4.2.3设p(亡)=(p,,(t))是全矩阵Q的一q一函数,G(t)代表相对应的Markov积分算子半群,其生成元用几表示.则下列陈述等价: (a)几在Ico空间里是稠定义的,也就是画两了=100; (b)q一矩阵Q是一致有界的,即s[lP‘。E吼<十、; (c)几是lco空间上的有界算子。 定理4.2.4设侧t)二恤,(t))是q-矩阵Q的一q一函数,G(约代表相对应的Malkov积分算子半群,其生成元用几表示。则下列陈述等价: (a)p。)二(p,,(艺))是Feller一Reuter--Riley转移函数; (b)几在c。空间里的部分几。生成。。空间上的正的强连续压缩半群。 同时若A是Feller-Reuter一形Icy转移函数尸(t)=恤j(t))在句空间上的生成元,则A二几。. 在Markov过程理论中,逼近问题一直是很重要的问题.通过离散骨架逼近具有可数无限状态空间的参数连续Markov链曾经被广泛的研究;william,David【lv]考虑了用状态空间有限的参数连续Markov链来逼近具有可数无限状态空间的参数连续Markov链;Anderson[2{证明了对Q一矩阵的最小转移函数,一定可以由一列具有一致有界Q一矩阵的转移函数来逼近.本文应用有界算子半群理论,从点态的角度研究转移函数的逼近,得到大量丰富的结果.在‘1空间上考虑时我们有如下结论: 定理5.1.2设Q=(、、,)是全矩阵,F。)=(几,(‘))是它的最小g一函数.设。Q=(。俄,)是。q一矩阵,。F(t)二(。人,(t))是它的最小。q一函数.如果满足如下条件: (a)入一Q在loo上是单射,也就是说对入>0,方程(久一Q)夕=0刀任100只有平凡解夕=。(b)当j尹乞时,,卿三钩,同时对任意的乞,J〔E,当。*co时,。卿。如那么当。。oo时,我们有。几(亡)仆五,(老)V乞,j任E,t全o定理5.1.3设Q二(卿)是q一矩阵,F(t)=(fij(t))是它的最小q一函数.假定(a)入一Q在l二上是单射,也就是说对人>0,方程(久一Q),=o,任100只有平凡解y=0; (b),。一矩阵。Q==(。钧)满足两个条件:(i)当饭,J<。时,取。。心=价s;(11)当‘全“或者jZ。时,。蜘在保证。Q=(。钧)是。伞矩阵的情况下可以任意取值. 最后,设。fij(t)是。Q二(n卿)的最小。q一函数.则当。。oo时,我们有。凡(t)。人,(t)V‘,j〔E,t全0· 关于转移函数(可以不是最小转移函数)P(约=帆了(t))与它对应的预解函数R(幻=(句(习)之间的连续依赖关系,我们有如下结论: 定理5.1.4设p(‘)==(p、,(t))和。p。)=(。尸、,(t))是转移函数,而R林)=(:*;(人))和。R(的=(。与(习)是它们相对应的预解函数.考虑如下陈述: (a)对任意的‘,j〔E和t全o,当n一co时,。尹、,(才)分p*j(t); (b)对任意的乞,j EE和入>0,当二。co时,。场(习。八,(的· 我们有:如果(a)成立,那么(b)一定成立;反过来,当。八,(劝三八,(习且(b)成立时,则(a)才成立. 接下来,在c。空间上考虑我们的问题.众所周知,Felkr一Reute卜Rilcy转移函数是句空间上的正的强连续压缩半群12].但是对于一给定的q一矩阵Q来说,其Feller-Reuter一几ley转移函数不一定存在(如果存在一定唯一,而且是最小q-函数Iz]),因此在句上空间考虑转移函数与它的q一矩阵Q之间的连续依赖关系时,我们必需事先假定它的Feller一取uter-Riley转移函数存在.这里要解决的问题是:对于一列点态收敛到q一矩阵Q的。q一矩阵。Q(n〔N),如果。。一矩阵。Q的最小转移函数。F(‘)=(。扬(t))是Feller一Reuter一Riley转移函数,那么是否。F(约二(。二,(t))收敛到q一矩阵Q的最小转移函数F(t)=(凡(t)),并且F(t)=(人,(t))也是Feller一Reuter-Riley转移函
【Abstract】 In the study of theories of Markov processes, there traditionally exist two methods: the probabilistic method and the analytical method. Recently mathematician investigate theories of Markov processes using the theory of semigroups of linear operators, and obtain many results. In this paper, we, mainly by means of analytic method and using the theory of semigroups of linear operators, study the transition functions, q-matrices, and the continuous dependence of transition functions on their q-matrices in continuous-time Markov chains (CTMCs).We know that every transition function is a positive strongly continuous semigroup of contractions on l1, but it isn’t a positive strongly continuous semigroup of contractions on l. Thus Anderson [2] think that l is too large a space on which to develop a really useful theory. However, are there similar applications of the integrated semigroups which have developed recently to CTMCs. Li Y.R. [33, 36] discussed the above question, and proved that the transition function is a positive once strong-contraction integrated semigroup on l(we call Markov integrated semigroup), and established the relationship between the transition functions and the Markov integrated semigroups. A natural question is: whether does there exist a subspace of l on which any transition function is a positive strongly continuous semigroup of contractions. According to the theory of Markov integrated semigroup developed by Li Y.R.[33, 36], we obtain the following results.Theorem 4.2.2 Let P(t) = (pij(t)) be a q-function of q-matrix Q, and G(t) be the Markov integrated semigroup with generator Q. DefineC1 = {, P(t)x is continuous function of t > 0 on l}Then the transition function P(t) is a positive contraction C0 semigroup on C1, and its generator 0 is the part of in C1, i.e.,Theorem 4.2.3 Let P(t) = (pij(t)) be a q-function of q-matrix Q, and G{t) be the Markov integrated semigroup with generator . Then the following statements are equivalent:(a) is densely defined in l, i.e.,D() = l(b) q-matrix Q is uniformly bounded, i.e., (c) is a bounded operator on Theorem 4.2.4 Let P(t) = (pij(t)) be a q-function of q-matrix Q, and G(t) be the Markov integrated semigroup with generator . Then the following statements are equivalent:(a) P(t) = (pij(t)) is a Feller-Reuter-Riley transition function;(b) the part of in c0 generates a positive C0 semigroup of contractions on c0.Moreover, let A be the generator of P{t) on cq. Then A is the part of in c0, i.e., A = 0.In the theories of Markov process, the question on approximation is always of most importance. Approximation by discrete skeletons has long been the mainstay of this theory. David Williams [17] considered a kind of approximation, that of the CTMCs by the finite CTMCs, where the finite CTMCs is well understood. Anderson [2] stated that there exists a sequence of transition functions with uniformly bounded q-matrices to converge to a given minimal transition function. In this paper, we study the same question by means of theory of semigroups of linear opreraors. Many useful results are obtained. Considering our question in l1, we have:Theorem 5.1.2 Let Q = (qij) be a g-matrix, and F(t) = (fij(t)) be the minimal q-function. Let nQ be nq-matrix, and nF{t) = {nfij(t)) be the minimal nq-fiinction. Suppose(a) - Q is injective on l; that is, the equationhas no solution other than the trivial solution y = 0 for some A > 0;ThenTheorem 5.1.3 Let Q = {qij) be a q--matrix, and F(t) = (fij(t)) be the minimal q-function. Suppose(a) - Q is injective on l for some A > 0;(b)The truncated matrix nQ = (nqij) satisfies (i) if i,j < n, nqij = qij; (ii) if other, nqij can be arbitrary to ensure that nQ is a nq-matrix.Finally, let nF(t) = {nfij{t)) be the minimal nq-function. Then,Theorem 5.1.4 Let P(t) = {pij(t)) and nP{t) = (nPij{t)) be transition functions (not necessary the minimal ones), and R(X) = (rij(A))and nR = (nrij be the corresponding resolvent functions, respectively. Consider the following statements:(a) nPij(t)
【Key words】 Continuous-time Markov chains; Transition functions; Feller-Reuter-Riley transition functions; Resolvent functions; (q-functions; q-matrices; Positive C0 semigroups of contractions; Markov integrated semigroups; Resolvent positive operators; Increasing integrated semigroups.;
- 【网络出版投稿人】 西南师范大学 【网络出版年期】2004年 04期
- 【分类号】O152.7
- 【下载频次】93