Reducing Markov Chains for Performance Evaluation
|
|
- Margery Simmons
- 5 years ago
- Views:
Transcription
1 1 Reducing Markov Chains for erformance Evaluation Y. ribadi, J..M. Voeten and B.D. Theelen Information and Communication Systems Group, Faculty of Electrical Engineering Eindhoven Embedded Systems Institute Eindhoven University of Technology.O.Box 513, 5600 MB Eindhoven, The Netherlands Abstract A major problem in performance evaluation of real-life industrial systems is the enormous number of states generated by the system s model during the simulation process. An example is in a arallel Object-Oriented Specification Language (OOSL) model, which can be interpreted as a discrete-time Markov chain enabling the evaluation of long-run average performance metrics. In general this Markov chains is huge, making performance simulation expensive since all performance estimations must be updated in each state. In this paper, we introduce a method for performance evaluation based on Markov chain theory, which aims at the reduction of the original Markov chain into a smaller chain. We prove that the reduction yields another Markov chain and it preserves several important properties. We further give a method to effectively construct the reduced chain. The reduction method yields an efficient way to estimate performance metrics during model simulation, since in general it requires estimations to be updated only in a relatively small part of the state space. Keywords Markov Chains; erformance Modelling; Ergodicity; Long-run Averages I. INTRODUCTION The OOSL language is a system level modeling language for complex real-time hardware/software systems [6], [3]. This language has proven to be very useful to describe real-life industrial systems and to verify their qualitative(correctness) and quantitative(performance) properties. In [6], [10] for instance, OOSL has been applied to design and analyze a distributed control system for a new generation of mailing machines. In [5], a protectionswitching protocol for SDH has been modeled, simulated and verified. A case study on the application of the OOSL formalism to analyze the performance properties of an industrial-sized Internet Router has been presented in [9]. The OOSL language is equipped with a lotkin-style structural operational semantics defining a labeled transi- This research is supported by ROGRESS, the embedded systems research program of the Dutch organization for Scientific Research NWO, the Dutch Ministry of Economic Affairs and the Technology Foundation STW. tion system. By resolving non-determinism with an external scheduler, this labeled transition system can be transformed into a discrete-time Markov chain [11]. erformance metrics can be expressed as long-run average rewards. A reward is a real-valued function on the state space of the chain and the idea is that each time a certain state is visited, a reward as specified by the reward function is obtained. Long-run average rewards can be computed using the ergodic theorem for Markov chains. However, the state space of Markov chains is often too large to perform such an exhaustive analysis. In case of simulation the size of the state space also yields a problem, since every performance estimation must be updated in each state, even if this state has no direct impact on the the estimation. For this reason, we develop a method to reduce the Markov chain by leaving out all such states. The plan of the remainder of this paper is as follows. In section II, we explain the formalism of Markov chains in terms of traces and their probabilities. In section III, long-run average rewards and the ergodic theorem are discussed. Section IV forms the core of this paper. In this section, we discuss the construction method and prove that it preserves several important properties. In particular, we prove that the long-run average reward of the original chain and that of the reduced chain are equal. We also show an example in this section. Finally, in Section V the conclusions are given. II. FORMALISM Assume a OOSL model defines a discrete time Markov chain { } with a countable state space S, an initial distribution D (where D { S } is a set of probability distributions) and a probability matrix S S. For S denotes the probability that. The def- the chain transits from state to state inition of such a chain will be given in the form of the ensemble S. Further, we assume that the probability space of this chain is given by probability triple F where is the collection of all infinite state se-
2 2 quences, F is the -algebra generated by the collection of cylindrical sets and is the corresponding probability measure (see [1] for more details). A finite state se- quence or infinite state sequence is called a trace if for each respectively each. For any infinite trace, we let denote the finite prefix of length and for any finite trace we define as the thin cylinder of all infinite traces having as prefix. The probabilities of the occurrences of finite trace and infinite trace are denoted by and respectively. These probabilities are given by where chain starts in state. (1) (2) is the probability that the Next to the probabilities and, we also need the probabilities of the occurrence of finite or infinite traces, conditional on the initial state of these traces. These conditional probabilities are given by and. Notice that and respectively. Let be a set of finite traces. Then is called proper if no trace in is a proper prefix of another trace in. A set is called initial if all traces in have the same initial state and final if all traces in For a proper set, we define by have the same final state. If is also initial, we define the conditional probability by Let be a proper final collection of finite traces and let be a proper initial collection of finite traces such that the final states in are the same as the initial states in. Then is defined by { and } Notice that is proper, and if is initial,. III. REWARDS AND ERFORMANCE EVALUATION For performance evaluation, the information contained in the states is used to define a number of rewards. A reward is a real-valued function on state space and the idea is that each time a certain state is visited a reward as specified by the reward function is obtained [7]. Definition 1: A reward is a function from state space S to R and for S, denotes the reward earned each time the Markov chain makes a transition into state. Definition 2: A reward function is called a conditional reward if or for all S. A state is called relevant (with respect to ) if and irrelevant (with respect to ) if. Next we define a reward structure Definition 3: Let { } be a Markov chain defined by S and let R be a set of reward functions. Then S R is called a reward structure. Many performance metrics can be expressed in terms of long-run average rewards or combination thereof. A reward structure defines precisely those rewards required to express these performance metrics. For reward and Markov chain { }, the long-run average -reward is defined by the limiting behaviour of random variable (3) With the limit of a random sequence we mean almost sure convergence [4] in this paper. Next, we define when the above limit exists. Definition 4: A Markov chain is called ergodic if it has a positive recurrent state that can be reached from any state with probability one. The property of ergodicity considered in this paper is taken from [2], [7]. The definition of ergodicity implies that the chain S has no two disjoint closed sets of states. Further, the (non-empty) collection of recurrent states is irreducible, each state in is positive recurrent and each state in \ is transient. It can be shown that an ergodic chain has a unique equilibrium distribution { S} where D. It further satisfies the following important ergodic theorem. Theorem 1 (Ergodic Theorem) Let { } be an ergodic Markov chain defined by a reward structure S R and let R be a reward function. Then (4) provided that converges absolutely.
3 roof: given in [2]. The proof is a generalization of the proof Definition 5: A reward structure S R is called complete if S is ergodic and if for each R, converges absolutely, where is the unique equilibrium distribution of the chain. Hence, in a complete reward structure, all long-run average rewards exists and can be computed using the ergodic theorem. We encountered that many performance metrics in practice can be expressed as conditional long-run average rewards of the form (see also [8]) where is any reward and where is a conditional reward. The idea of the performance metric above is that reward is averaged precisely over those states that are relevant with respect to. It is easy to prove that if and then the performance metric ). in (5) converges almost surely to (if IV. REDUCING MARKOV CHAINS The conditional reward defined in (5) can be computed from the original Markov chain obtained from the OOSL model. In case of simulation, the estimation of (5) has to be updated in each simulation step, which is computationally expensive. Now, since the conditional long-run average reward (5) only depends on the relevant states in the initial chain, it might seem plausible that we only have to consider these relevant states in order to compute or estimate the metric. In this paper, we will show that this is indeed the case. To this end, we will reduce the original chain by leaving out all irrelevant states (with respect to ) and prove that the conditional long-run average -reward in the original chain is the same as the long-run average -reward in the reduced chain. We now fix a complete reward structure S R and assume that it defines Markov chain { } which is based on probability triple F. We further assume that R is a reward and R is a conditional reward. In this section we will define a reduced reward structure (with respect to ) which will be denoted by S R = S R. Here the state space S is defined by S { S } and R is the collection of rewards in R reduced to the states in S. To define components and we first introduce the reduced Markov chain denoted by { }. This reduced chain will be defined on the same probability (5) triple F as the original chain. For each and infinite trace, is defined by the relevant state in undefined if it exists otherwise Notice that { } defined this way is not always a stochastic process. In general, is not even a function. However, the following lemma states that if the original Markov chain has at least one positive recurrent relevant state, then { } defines a proper stochastic process. Lemma 1: Let the original Markov chain { } has at least one positive recurrent relevant state. { 3 Then } is a stochastic process. Assume state is a positive recurrent rel- roof: evant state. Then each infinite trace has infinitely many occurrences of state with probability one. Hence, if we discard a set of probability zero [2], is a complete S -valued function. We further have to show that each is a random variable. But this follows easily from the fact that each set of infinite traces is measurable. Next, we want to prove that the stochastic process { } is in fact a Markov chain. Theorem 2: Let the original Markov chain { } have at least one positive recurrent relevant state. Then { } is a Markov chain. To prove this theorem, we first introduce the following collections of finite traces. For S and S, we define = The set of finite traces that end in and for which all preceding states are irrelevant = The set of finite traces that start in, end in and has no relevant states in between Notice that is an initial and final proper set of traces and that is a final proper set of traces. We are now able to prove theorem 2. roof: We will first prove the Markovian property.
4 4 Similarly it is not hard to show that and hence the Markovian property holds. The property of time homogeneity is proven in an analogous way. From the previous proof, it follows immediately that the probability that the reduced chain transits from S to S is given by. Hence the reduced probability matrix is defined by for all S Further, the initial distribution of the reduced chain is defined by for all S So now we have defined the reduced reward structure S R where S defines the reduced Markov chain { }. In the next subsection IV-A, we will show that the reduced reward structure is ergodic if the original chain has at least one positive recurrent relevant state. We further prove that the reduced reward structure is in fact complete. In subsection IV-B, we will prove the preservation of long-run averages. In subsection IV-C, a method to effectively construct the reduced chain is given. Finally, an example is given in subsection IV-D. A. reservation of Ergodicity and Completeness For the preservation of ergodicity, we need to prove the following theorem. Theorem 3: If the original chain { } is ergodic and has a positive recurrent relevant state, then the reduced Markov chain { } is ergodic. roof: Assume that the original chain is ergodic and let be a positive recurrent relevant state. We will show that is a positive recurrent state that is reached from any initial state in the reduced chain with probability one. Let S be any initial state of the reduced chain and define as the set of finite traces that start in, end in and have no other occurrences of other than the final state. We have to prove that. But this follows immediately from the fact that is reached with probability one in the original chain. To show that is a recurrent state, define as the set of finite traces, consisting of at least two states, that start in, end in and have no other occurrences of state in between. Since is a recurrent state in the original chain, and hence is a recurrent state in the reduced chain. To prove that is positive recurrent, we need to show that the mean return time to is finite. To this end, define for trace, as the amount of states in minus one and as the amount of relevant states in minus one. We have to show that (6) But since is a positive state in the original chain, we clearly have (7) from which the result follows. For the preservation of completeness, we need to prove the following theorem. Theorem 4: If the original reward structure S has a positive recurrent relevant state, then the reduced reward structure S roof: R is complete. By the previous theorem we already know that the reduced chain is ergodic so that the unique equilibrium distribution D exists. We further have to show that for each R the following series converges absolutely R (8) By [2] this is equivalent by saying that the expected cumulative absolute reward earned during one cycle through a positive recurrent relevant state is finite. By an argument similar to that in (7) it can be verified that this condition is indeed satisfied. B. reservation of Long-run Averages For the preservation of long-run averages, we need to prove the following main result. Theorem 5: Assume the original reward structure S R is complete and has a positive recurrent relevant state. Then roof: iff To show this, let be a constant such that. Next, we define a set }. We further { define set { }. For the only if part we have to prove that
5 implies. To this end, assume and let. Then we have that. By the definition of it then easily follows that ( the relevant state in ), so. Hence and. The if part is proven in a similar way. Notice that in case the condition of the theorem is satisfied, both and converge almost surely, where the latter limit is different from 0. The next corollary establishes the relation between the equilibrium distribution of the original chain and that of the reduced one. Corollary 1: Assume that the original Markov chain has a positive recurrent relevant state. Let and denote the equilibrium distributions of { } and { } respectively. Then for each S (9) roof: Let S be any relevant state. We define a reward as the indicator function of state, i.e. yielding 1 in state and 0 in any other state in S. Now although is not necessary in R, it is easy to see that Theorem 5 also holds for this reward. Hence, by the ergodic theorem we obtain Now by observing that for all, =, the result follows straightforwardly. C. Construction Method Although the reduced reward structure has been defined already in the beginning of Section IV, this definition is not constructive. Computing the transition probability from relevant state to relevant state as the conditional probability of set or computing the initial probability of relevant state as the probability of set is a laborious exercise. In this subsection, we introduce a method for computing the transition probabilities by solving a set of linear equations. The initial probabilities can then be computed immediately from the initial distribution of the original chain and the probability matrix of the reduced chain. Theorem 6: Assume the original complete reward structure has a positive recurrent relevant state. Extent the reduced probability matrix to S S and define for all S S Then the elements of satisfy the following set of linear equations: \S roof: It is easily verified that for S and S the following equation holds { } { } \S By working out the conditional probability of, the result follows. We have strong evidence that the system of linear equations has a unique solution, but more research is necessary to establish a formal proof. Theorem 7: Assume the original complete reward structure has a positive recurrent relevant state. Then the elements of satisfy the following equations: 6. \S roof: The proof is similar to the proof of Theorem D. Example We give an example to illustrate what we have discussed so far. Suppose a OOSL model defines a Markov chain with transition probabilities as given in Fig. 1 (a). The realvalued reward is indicated in each state. Relevant states are indicated with a black dot. We assume that the initial distribution is given by,, = = =. Using the ergodic theorem, we can calculate the equilibrium distribution. It is given by,, = = =. Using the construction method, the reduced Markov chain shown in Fig. 1 (b) is obtained. Using Theorem 6, the reduced transition probabilities are computed and by Theorem 7 the reduced initial distribution is calculated. This distribution is given by and for all the other relevant states. Using the ergodic theorem or by applying Corollary 1, the reduced equilibrium distribution can be obtained easily. This distribution is given by, and. The reader may want to check that the conditional longrun average -reward in the original Markov chain is the same as the long-run average reward in the reduced Markov chain, as is stated by Theorem 5. Both long-run averages equal 2. It is clear that in case of simulation, the reduced Markov chain is preferred above the original Markov chain to estimate the metric. Notice that in this case the original chain 5
6 6 Fig. 1. (a) Original Markov chain and (b) Reduced Markov chain can be used to generate a trace for analysis; it is not necessary to first construct explicitly the reduced chain 1. During the analysis of this trace, only the relevant states are observed. Since in practice most states will be irrelevant for the performance metric in question, the reduction technique is expected to result in a huge increase in simulation performance. To be able to apply simulation of the reduced chain, the central limit theorem for Markov chains should hold. It is expected that the construction method preserves the conditions for this limit theorem to hold, but this will be a topic for future research. V. CONCLUSIONS In this paper, we introduced a reduction method for Markov chains. The reduction consists of leaving out all states not satisfying the update condition specified by some conditional reward. We have proved that the reduction yields another Markov chain and that it preserves the property of ergodicity. We further showed that every conditional long-run average reward in the original chain is equal to the long-run average reward in the reduced chain. We gave a method to effectively construct the reduced Markov chain. The presented reduction method is especially useful in the area of simulation and performance estimation. The reduction method makes that performance estimations only Notice that the reduced chain is defined using the original chain. -algebra of the have to be updated in those states that satisfy the update condition specified by some conditional reward. Since in practice most states will not satisfy this reward, the reduction technique is expected to result in a huge increase in simulation performance. REFERENCES [1]. Billingsley. robability and Measure. Wiley, New York, [2] K.L. Chung. Markov Chains with Stationary Transition robabilities, Springer-Verlag, Berlin, [3] M.C.W. Geilen and J..M. Voeten. Real-time Concepts for a Formal Specification Language for Software/Hardware Systems. In: J.. Veen, Ed., roceeding of rorisc 97, pp Utrecht(Netherlands): STW, Technology Foundation, [4] A.F. Karr. robability. Springer-Verlang, New York, [5] G. Lopez. Modelisation, Simulation et Verification d un rotocole de Telecommunication. Graduation Report. Eindhoven University of Technology, Eindhoven (Netherlands), [6].H.A. van der utten and J..M. Voeten. Specification of Reactive Hardware/Software Systems. h.d. Thesis. Eindhoven(Netherlands), Eindhoven University of Technology, [7] H.C. Tijms. Stochastic Models-An Algorithmic Approach. John Wiley & Sons, Chichester, England, [8] B.D. Theelen, J..M. Voeten and Y. ribadi. Accuracy Analysis of Long-Run Average erformance Metrics. In: roceeding of ROGRESS 01. Utrecht(The Netherlands): STW Technology Foundation, [9] B.D. Theelen, J..M. Voeten, L.J. van Bokhoven, G.G. de Jong, A.M.M. Niemegeers,.H.A. van der utten, M..J. Stevens, J.C.M. Baeten. System-level Modeling and erformance Analysis. In: J.. Veen, Ed., roceedings of ROGRESS 00, pp Utrecht(The Netherlands): STW Technology Foundation, [10] J..M. Voeten,.H.A. van der utten and M..J. Stevens. Sys-
7 tematic Development of Industrial Control Systems Using Software/Hardware Engineering. In:. Milligan and. Corr, Eds., roc. of EUROMICRO 97, pp Los Alamitos, California(U.S.A): IEEE, [11] J..M. Voeten. erformance Evaluation with Temporal Rewards. To be published in: Journal of erformance Evaluation,
Summary of Results on Markov Chains. Abstract
Summary of Results on Markov Chains Enrico Scalas 1, 1 Laboratory on Complex Systems. Dipartimento di Scienze e Tecnologie Avanzate, Università del Piemonte Orientale Amedeo Avogadro, Via Bellini 25 G,
More informationMarkov chains. Randomness and Computation. Markov chains. Markov processes
Markov chains Randomness and Computation or, Randomized Algorithms Mary Cryan School of Informatics University of Edinburgh Definition (Definition 7) A discrete-time stochastic process on the state space
More informationRegenerative Processes. Maria Vlasiou. June 25, 2018
Regenerative Processes Maria Vlasiou June 25, 218 arxiv:144.563v1 [math.pr] 22 Apr 214 Abstract We review the theory of regenerative processes, which are processes that can be intuitively seen as comprising
More informationErgodic Theorems. Samy Tindel. Purdue University. Probability Theory 2 - MA 539. Taken from Probability: Theory and examples by R.
Ergodic Theorems Samy Tindel Purdue University Probability Theory 2 - MA 539 Taken from Probability: Theory and examples by R. Durrett Samy T. Ergodic theorems Probability Theory 1 / 92 Outline 1 Definitions
More informationIN THIS paper we investigate the diagnosability of stochastic
476 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL 50, NO 4, APRIL 2005 Diagnosability of Stochastic Discrete-Event Systems David Thorsley and Demosthenis Teneketzis, Fellow, IEEE Abstract We investigate
More informationLecture 20 : Markov Chains
CSCI 3560 Probability and Computing Instructor: Bogdan Chlebus Lecture 0 : Markov Chains We consider stochastic processes. A process represents a system that evolves through incremental changes called
More informationChapter 2: Markov Chains and Queues in Discrete Time
Chapter 2: Markov Chains and Queues in Discrete Time L. Breuer University of Kent 1 Definition Let X n with n N 0 denote random variables on a discrete space E. The sequence X = (X n : n N 0 ) is called
More informationMarkov Chains, Stochastic Processes, and Matrix Decompositions
Markov Chains, Stochastic Processes, and Matrix Decompositions 5 May 2014 Outline 1 Markov Chains Outline 1 Markov Chains 2 Introduction Perron-Frobenius Matrix Decompositions and Markov Chains Spectral
More informationFormal Models for Encapsulation, Structure and Hierarchy in Distributed Systems
Formal Models for Encapsulation, Structure and Hierarchy in Distributed Systems M.C.W. Geilen Section of Information and Communication Systems, Faculty of Electrical Engineering, Eindhoven University of
More informationAn On-the-fly Tableau Construction for a Real-Time Temporal Logic
#! & F $ F ' F " F % An On-the-fly Tableau Construction for a Real-Time Temporal Logic Marc Geilen and Dennis Dams Faculty of Electrical Engineering, Eindhoven University of Technology P.O.Box 513, 5600
More informationLecture 5. If we interpret the index n 0 as time, then a Markov chain simply requires that the future depends only on the present and not on the past.
1 Markov chain: definition Lecture 5 Definition 1.1 Markov chain] A sequence of random variables (X n ) n 0 taking values in a measurable state space (S, S) is called a (discrete time) Markov chain, if
More informationSection 2: Classes of Sets
Section 2: Classes of Sets Notation: If A, B are subsets of X, then A \ B denotes the set difference, A \ B = {x A : x B}. A B denotes the symmetric difference. A B = (A \ B) (B \ A) = (A B) \ (A B). Remarks
More informationWeek 5: Markov chains Random access in communication networks Solutions
Week 5: Markov chains Random access in communication networks Solutions A Markov chain model. The model described in the homework defines the following probabilities: P [a terminal receives a packet in
More informationThe State Explosion Problem
The State Explosion Problem Martin Kot August 16, 2003 1 Introduction One from main approaches to checking correctness of a concurrent system are state space methods. They are suitable for automatic analysis
More informationLecture 10. Theorem 1.1 [Ergodicity and extremality] A probability measure µ on (Ω, F) is ergodic for T if and only if it is an extremal point in M.
Lecture 10 1 Ergodic decomposition of invariant measures Let T : (Ω, F) (Ω, F) be measurable, and let M denote the space of T -invariant probability measures on (Ω, F). Then M is a convex set, although
More informationLöwenheim-Skolem Theorems, Countable Approximations, and L ω. David W. Kueker (Lecture Notes, Fall 2007)
Löwenheim-Skolem Theorems, Countable Approximations, and L ω 0. Introduction David W. Kueker (Lecture Notes, Fall 2007) In its simplest form the Löwenheim-Skolem Theorem for L ω1 ω states that if σ L ω1
More informationReinforcement Learning
Reinforcement Learning March May, 2013 Schedule Update Introduction 03/13/2015 (10:15-12:15) Sala conferenze MDPs 03/18/2015 (10:15-12:15) Sala conferenze Solving MDPs 03/20/2015 (10:15-12:15) Aula Alpha
More information25.1 Ergodicity and Metric Transitivity
Chapter 25 Ergodicity This lecture explains what it means for a process to be ergodic or metrically transitive, gives a few characterizes of these properties (especially for AMS processes), and deduces
More informationFirst-order resolution for CTL
First-order resolution for Lan Zhang, Ullrich Hustadt and Clare Dixon Department of Computer Science, University of Liverpool Liverpool, L69 3BX, UK {Lan.Zhang, U.Hustadt, CLDixon}@liverpool.ac.uk Abstract
More informationStatistics 992 Continuous-time Markov Chains Spring 2004
Summary Continuous-time finite-state-space Markov chains are stochastic processes that are widely used to model the process of nucleotide substitution. This chapter aims to present much of the mathematics
More informationGeneral Glivenko-Cantelli theorems
The ISI s Journal for the Rapid Dissemination of Statistics Research (wileyonlinelibrary.com) DOI: 10.100X/sta.0000......................................................................................................
More informationChapter 2 SOME ANALYTICAL TOOLS USED IN THE THESIS
Chapter 2 SOME ANALYTICAL TOOLS USED IN THE THESIS 63 2.1 Introduction In this chapter we describe the analytical tools used in this thesis. They are Markov Decision Processes(MDP), Markov Renewal process
More informationTopics in Stochastic Geometry. Lecture 4 The Boolean model
Institut für Stochastik Karlsruher Institut für Technologie Topics in Stochastic Geometry Lecture 4 The Boolean model Lectures presented at the Department of Mathematical Sciences University of Bath May
More informationSTOCHASTIC PROCESSES Basic notions
J. Virtamo 38.3143 Queueing Theory / Stochastic processes 1 STOCHASTIC PROCESSES Basic notions Often the systems we consider evolve in time and we are interested in their dynamic behaviour, usually involving
More informationLet (Ω, F) be a measureable space. A filtration in discrete time is a sequence of. F s F t
2.2 Filtrations Let (Ω, F) be a measureable space. A filtration in discrete time is a sequence of σ algebras {F t } such that F t F and F t F t+1 for all t = 0, 1,.... In continuous time, the second condition
More informationP i [B k ] = lim. n=1 p(n) ii <. n=1. V i :=
2.7. Recurrence and transience Consider a Markov chain {X n : n N 0 } on state space E with transition matrix P. Definition 2.7.1. A state i E is called recurrent if P i [X n = i for infinitely many n]
More informationThe cost/reward formula has two specific widely used applications:
Applications of Absorption Probability and Accumulated Cost/Reward Formulas for FDMC Friday, October 21, 2011 2:28 PM No class next week. No office hours either. Next class will be 11/01. The cost/reward
More informationThe efficiency of identifying timed automata and the power of clocks
The efficiency of identifying timed automata and the power of clocks Sicco Verwer a,b,1,, Mathijs de Weerdt b, Cees Witteveen b a Eindhoven University of Technology, Department of Mathematics and Computer
More informationMATRIX LIE GROUPS AND LIE GROUPS
MATRIX LIE GROUPS AND LIE GROUPS Steven Sy December 7, 2005 I MATRIX LIE GROUPS Definition: A matrix Lie group is a closed subgroup of Thus if is any sequence of matrices in, and for some, then either
More informationMath Homework 5 Solutions
Math 45 - Homework 5 Solutions. Exercise.3., textbook. The stochastic matrix for the gambler problem has the following form, where the states are ordered as (,, 4, 6, 8, ): P = The corresponding diagram
More informationN.G.Bean, D.A.Green and P.G.Taylor. University of Adelaide. Adelaide. Abstract. process of an MMPP/M/1 queue is not a MAP unless the queue is a
WHEN IS A MAP POISSON N.G.Bean, D.A.Green and P.G.Taylor Department of Applied Mathematics University of Adelaide Adelaide 55 Abstract In a recent paper, Olivier and Walrand (994) claimed that the departure
More informationLecture 7. µ(x)f(x). When µ is a probability measure, we say µ is a stationary distribution.
Lecture 7 1 Stationary measures of a Markov chain We now study the long time behavior of a Markov Chain: in particular, the existence and uniqueness of stationary measures, and the convergence of the distribution
More informationMarkov Chains CK eqns Classes Hitting times Rec./trans. Strong Markov Stat. distr. Reversibility * Markov Chains
Markov Chains A random process X is a family {X t : t T } of random variables indexed by some set T. When T = {0, 1, 2,... } one speaks about a discrete-time process, for T = R or T = [0, ) one has a continuous-time
More informationMarkov Chains (Part 4)
Markov Chains (Part 4) Steady State Probabilities and First Passage Times Markov Chains - 1 Steady-State Probabilities Remember, for the inventory example we had (8) P &.286 =.286.286 %.286 For an irreducible
More information2. Transience and Recurrence
Virtual Laboratories > 15. Markov Chains > 1 2 3 4 5 6 7 8 9 10 11 12 2. Transience and Recurrence The study of Markov chains, particularly the limiting behavior, depends critically on the random times
More informationStochastic processes. MAS275 Probability Modelling. Introduction and Markov chains. Continuous time. Markov property
Chapter 1: and Markov chains Stochastic processes We study stochastic processes, which are families of random variables describing the evolution of a quantity with time. In some situations, we can treat
More informationSMSTC (2007/08) Probability.
SMSTC (27/8) Probability www.smstc.ac.uk Contents 12 Markov chains in continuous time 12 1 12.1 Markov property and the Kolmogorov equations.................... 12 2 12.1.1 Finite state space.................................
More informationJOSEPH B. KADANE AND JIASHUN JIN, IN HONOR OF ARNOLD ZELLNER
UNIFORM DISTRIBUTIONS ON THE INTEGERS: A CONNECTION TO THE BERNOUILLI RANDOM WALK JOSEPH B. KADANE AND JIASHUN JIN, IN HONOR OF ARNOLD ZELLNER Abstract. Associate to each subset of the integers its almost
More informationDiscrete-Time Markov Decision Processes
CHAPTER 6 Discrete-Time Markov Decision Processes 6.0 INTRODUCTION In the previous chapters we saw that in the analysis of many operational systems the concepts of a state of a system and a state transition
More informationInventory Ordering Control for a Retrial Service Facility System Semi- MDP
International Journal of Engineering Science Invention (IJESI) ISS (Online): 239 6734, ISS (Print): 239 6726 Volume 7 Issue 6 Ver I June 208 PP 4-20 Inventory Ordering Control for a Retrial Service Facility
More informationLecture 3: Markov chains.
1 BIOINFORMATIK II PROBABILITY & STATISTICS Summer semester 2008 The University of Zürich and ETH Zürich Lecture 3: Markov chains. Prof. Andrew Barbour Dr. Nicolas Pétrélis Adapted from a course by Dr.
More informationMarkov Decision Processes with Multiple Long-run Average Objectives
Markov Decision Processes with Multiple Long-run Average Objectives Krishnendu Chatterjee UC Berkeley c krish@eecs.berkeley.edu Abstract. We consider Markov decision processes (MDPs) with multiple long-run
More informationMarkov Chains. X(t) is a Markov Process if, for arbitrary times t 1 < t 2 <... < t k < t k+1. If X(t) is discrete-valued. If X(t) is continuous-valued
Markov Chains X(t) is a Markov Process if, for arbitrary times t 1 < t 2
More informationEmbedded systems specification and design
Embedded systems specification and design David Kendall David Kendall Embedded systems specification and design 1 / 21 Introduction Finite state machines (FSM) FSMs and Labelled Transition Systems FSMs
More informationMarkov Chains (Part 3)
Markov Chains (Part 3) State Classification Markov Chains - State Classification Accessibility State j is accessible from state i if p ij (n) > for some n>=, meaning that starting at state i, there is
More informationStochastic Models: Markov Chains and their Generalizations
Scuola di Dottorato in Scienza ed Alta Tecnologia Dottorato in Informatica Universita di Torino Stochastic Models: Markov Chains and their Generalizations Gianfranco Balbo e Andras Horvath Outline Introduction
More informationSIMILAR MARKOV CHAINS
SIMILAR MARKOV CHAINS by Phil Pollett The University of Queensland MAIN REFERENCES Convergence of Markov transition probabilities and their spectral properties 1. Vere-Jones, D. Geometric ergodicity in
More informationThe matrix approach for abstract argumentation frameworks
The matrix approach for abstract argumentation frameworks Claudette CAYROL, Yuming XU IRIT Report RR- -2015-01- -FR February 2015 Abstract The matrices and the operation of dual interchange are introduced
More informationProbabilistic Juggling
Probabilistic Juggling Arvind Ayyer (joint work with Jeremie Bouttier, Sylvie Corteel and Francois Nunzi) arxiv:1402:3752 Conference on STOCHASTIC SYSTEMS AND APPLICATIONS (aka Borkarfest) Indian Institute
More informationA Scenario-Aware Data Flow Model for Combined Long-Run Average and Worst-Case Performance Analysis
A Scenario-Aware Data Flow Model for Combined Long-Run Average and Worst-Case Performance Analysis B.D. Theelen 1, M.C.W. Geilen 1, T. Basten 1, J.P.M. Voeten 1,2, S.V. Gheorghita 1 and S. Stuijk 1 1 Eindhoven
More informationPropositional and Predicate Logic - V
Propositional and Predicate Logic - V Petr Gregor KTIML MFF UK WS 2016/2017 Petr Gregor (KTIML MFF UK) Propositional and Predicate Logic - V WS 2016/2017 1 / 21 Formal proof systems Hilbert s calculus
More informationSimilarities between Timing Constraints
Similarities between Timing Constraints Towards Interchangeable Constraint Models for Real-World Software Systems yyu8@iit.edu Similarities between Timing Constraints p. 1/11 Background Software for real-world
More informationUsing Markov Chains To Model Human Migration in a Network Equilibrium Framework
Using Markov Chains To Model Human Migration in a Network Equilibrium Framework Jie Pan Department of Mathematics and Computer Science Saint Joseph s University Philadelphia, PA 19131 Anna Nagurney School
More information8. Statistical Equilibrium and Classification of States: Discrete Time Markov Chains
8. Statistical Equilibrium and Classification of States: Discrete Time Markov Chains 8.1 Review 8.2 Statistical Equilibrium 8.3 Two-State Markov Chain 8.4 Existence of P ( ) 8.5 Classification of States
More informationStochastic Processes (Week 6)
Stochastic Processes (Week 6) October 30th, 2014 1 Discrete-time Finite Markov Chains 2 Countable Markov Chains 3 Continuous-Time Markov Chains 3.1 Poisson Process 3.2 Finite State Space 3.2.1 Kolmogrov
More informationComputing Consecutive-Type Reliabilities Non-Recursively
IEEE TRANSACTIONS ON RELIABILITY, VOL. 52, NO. 3, SEPTEMBER 2003 367 Computing Consecutive-Type Reliabilities Non-Recursively Galit Shmueli Abstract The reliability of consecutive-type systems has been
More informationOn Equilibria of Distributed Message-Passing Games
On Equilibria of Distributed Message-Passing Games Concetta Pilotto and K. Mani Chandy California Institute of Technology, Computer Science Department 1200 E. California Blvd. MC 256-80 Pasadena, US {pilotto,mani}@cs.caltech.edu
More informationMATH 56A SPRING 2008 STOCHASTIC PROCESSES 65
MATH 56A SPRING 2008 STOCHASTIC PROCESSES 65 2.2.5. proof of extinction lemma. The proof of Lemma 2.3 is just like the proof of the lemma I did on Wednesday. It goes like this. Suppose that â is the smallest
More information5 Set Operations, Functions, and Counting
5 Set Operations, Functions, and Counting Let N denote the positive integers, N 0 := N {0} be the non-negative integers and Z = N 0 ( N) the positive and negative integers including 0, Q the rational numbers,
More information1 Completeness Theorem for First Order Logic
1 Completeness Theorem for First Order Logic There are many proofs of the Completeness Theorem for First Order Logic. We follow here a version of Henkin s proof, as presented in the Handbook of Mathematical
More informationNonparametric inference for ergodic, stationary time series.
G. Morvai, S. Yakowitz, and L. Györfi: Nonparametric inference for ergodic, stationary time series. Ann. Statist. 24 (1996), no. 1, 370 379. Abstract The setting is a stationary, ergodic time series. The
More informationModel Checking. Temporal Logic. Fifth International Symposium in Programming, volume. of concurrent systems in CESAR. In Proceedings of the
Sérgio Campos, Edmund Why? Advantages: No proofs Fast Counter-examples No problem with partial specifications can easily express many concurrency properties Main Disadvantage: State Explosion Problem Too
More informationLimiting Behavior of Markov Chains with Eager Attractors
Limiting Behavior of Markov Chains with Eager Attractors Parosh Aziz Abdulla Uppsala University, Sweden. parosh@it.uu.se Noomene Ben Henda Uppsala University, Sweden. Noomene.BenHenda@it.uu.se Sven Sandberg
More informationQuantitative Verification
Quantitative Verification Chapter 3: Markov chains Jan Křetínský Technical University of Munich Winter 207/8 / 84 Motivation 2 / 84 Example: Simulation of a die by coins Knuth & Yao die Simulating a Fair
More informationComputer-Aided Program Design
Computer-Aided Program Design Spring 2015, Rice University Unit 3 Swarat Chaudhuri February 5, 2015 Temporal logic Propositional logic is a good language for describing properties of program states. However,
More informationAN EXPLORATION OF THE METRIZABILITY OF TOPOLOGICAL SPACES
AN EXPLORATION OF THE METRIZABILITY OF TOPOLOGICAL SPACES DUSTIN HEDMARK Abstract. A study of the conditions under which a topological space is metrizable, concluding with a proof of the Nagata Smirnov
More informationPredictable real-time software synthesis
Real-Time Syst (2007) 36: 159 198 DOI 10.1007/s11241-007-9013-6 Predictable real-time software synthesis Jinfeng Huang Jeroen Voeten Henk Corporaal Published online: 28 March 2007 Springer Science+Business
More informationSTA 624 Practice Exam 2 Applied Stochastic Processes Spring, 2008
Name STA 624 Practice Exam 2 Applied Stochastic Processes Spring, 2008 There are five questions on this test. DO use calculators if you need them. And then a miracle occurs is not a valid answer. There
More informationBias-Variance Error Bounds for Temporal Difference Updates
Bias-Variance Bounds for Temporal Difference Updates Michael Kearns AT&T Labs mkearns@research.att.com Satinder Singh AT&T Labs baveja@research.att.com Abstract We give the first rigorous upper bounds
More informationNon-homogeneous random walks on a semi-infinite strip
Non-homogeneous random walks on a semi-infinite strip Chak Hei Lo Joint work with Andrew R. Wade World Congress in Probability and Statistics 11th July, 2016 Outline Motivation: Lamperti s problem Our
More informationDefinition A finite Markov chain is a memoryless homogeneous discrete stochastic process with a finite number of states.
Chapter 8 Finite Markov Chains A discrete system is characterized by a set V of states and transitions between the states. V is referred to as the state space. We think of the transitions as occurring
More informationEstimation and Approximation Bounds for Gradient-Based Reinforcement Learning
Estimation and pproximation Bounds for radient-based Reinforcement Learning eter L. Bartlett and onathan Baxter Research School of Information Sciences and Engineering ustralian National University Canberra
More informationOn Finding Optimal Policies for Markovian Decision Processes Using Simulation
On Finding Optimal Policies for Markovian Decision Processes Using Simulation Apostolos N. Burnetas Case Western Reserve University Michael N. Katehakis Rutgers University February 1995 Abstract A simulation
More informationVerification of Probabilistic Systems with Faulty Communication
Verification of Probabilistic Systems with Faulty Communication P. A. Abdulla 1, N. Bertrand 2, A. Rabinovich 3, and Ph. Schnoebelen 2 1 Uppsala University, Sweden 2 LSV, ENS de Cachan, France 3 Tel Aviv
More informationMarkovian Description of Irreversible Processes and the Time Randomization (*).
Markovian Description of Irreversible Processes and the Time Randomization (*). A. TRZĘSOWSKI and S. PIEKARSKI Institute of Fundamental Technological Research, Polish Academy of Sciences ul. Świętokrzyska
More informationDiscrete time Markov chains. Discrete Time Markov Chains, Limiting. Limiting Distribution and Classification. Regular Transition Probability Matrices
Discrete time Markov chains Discrete Time Markov Chains, Limiting Distribution and Classification DTU Informatics 02407 Stochastic Processes 3, September 9 207 Today: Discrete time Markov chains - invariant
More informationNew Complexity Results for Some Linear Counting Problems Using Minimal Solutions to Linear Diophantine Equations
New Complexity Results for Some Linear Counting Problems Using Minimal Solutions to Linear Diophantine Equations (Extended Abstract) Gaoyan Xie, Cheng Li and Zhe Dang School of Electrical Engineering and
More informationProbabilistic Model Checking and Strategy Synthesis for Robot Navigation
Probabilistic Model Checking and Strategy Synthesis for Robot Navigation Dave Parker University of Birmingham (joint work with Bruno Lacerda, Nick Hawes) AIMS CDT, Oxford, May 2015 Overview Probabilistic
More informationRandom Times and Their Properties
Chapter 6 Random Times and Their Properties Section 6.1 recalls the definition of a filtration (a growing collection of σ-fields) and of stopping times (basically, measurable random times). Section 6.2
More informationA comment on Boucherie product-form results
A comment on Boucherie product-form results Andrea Marin Dipartimento di Informatica Università Ca Foscari di Venezia Via Torino 155, 30172 Venezia Mestre, Italy {balsamo,marin}@dsi.unive.it Abstract.
More informationInformation Theory and Statistics Lecture 3: Stationary ergodic processes
Information Theory and Statistics Lecture 3: Stationary ergodic processes Łukasz Dębowski ldebowsk@ipipan.waw.pl Ph. D. Programme 2013/2014 Measurable space Definition (measurable space) Measurable space
More informationLIMITS FOR QUEUES AS THE WAITING ROOM GROWS. Bell Communications Research AT&T Bell Laboratories Red Bank, NJ Murray Hill, NJ 07974
LIMITS FOR QUEUES AS THE WAITING ROOM GROWS by Daniel P. Heyman Ward Whitt Bell Communications Research AT&T Bell Laboratories Red Bank, NJ 07701 Murray Hill, NJ 07974 May 11, 1988 ABSTRACT We study the
More informationJónsson posets and unary Jónsson algebras
Jónsson posets and unary Jónsson algebras Keith A. Kearnes and Greg Oman Abstract. We show that if P is an infinite poset whose proper order ideals have cardinality strictly less than P, and κ is a cardinal
More informationMarkov Chains Handout for Stat 110
Markov Chains Handout for Stat 0 Prof. Joe Blitzstein (Harvard Statistics Department) Introduction Markov chains were first introduced in 906 by Andrey Markov, with the goal of showing that the Law of
More informationPermutation Excess Entropy and Mutual Information between the Past and Future
Permutation Excess Entropy and Mutual Information between the Past and Future Taichi Haruna 1, Kohei Nakajima 2 1 Department of Earth & Planetary Sciences, Graduate School of Science, Kobe University,
More informationStochastic Processes. Theory for Applications. Robert G. Gallager CAMBRIDGE UNIVERSITY PRESS
Stochastic Processes Theory for Applications Robert G. Gallager CAMBRIDGE UNIVERSITY PRESS Contents Preface page xv Swgg&sfzoMj ybr zmjfr%cforj owf fmdy xix Acknowledgements xxi 1 Introduction and review
More information1 Basic concepts from probability theory
Basic concepts from probability theory This chapter is devoted to some basic concepts from probability theory.. Random variable Random variables are denoted by capitals, X, Y, etc. The expected value or
More informationBasic Definitions: Indexed Collections and Random Functions
Chapter 1 Basic Definitions: Indexed Collections and Random Functions Section 1.1 introduces stochastic processes as indexed collections of random variables. Section 1.2 builds the necessary machinery
More informationS-adic sequences A bridge between dynamics, arithmetic, and geometry
S-adic sequences A bridge between dynamics, arithmetic, and geometry J. M. Thuswaldner (joint work with P. Arnoux, V. Berthé, M. Minervino, and W. Steiner) Marseille, November 2017 PART 3 S-adic Rauzy
More informationStability of the two queue system
Stability of the two queue system Iain M. MacPhee and Lisa J. Müller University of Durham Department of Mathematical Science Durham, DH1 3LE, UK (e-mail: i.m.macphee@durham.ac.uk, l.j.muller@durham.ac.uk)
More informationFuzzy Limits of Functions
Fuzzy Limits of Functions Mark Burgin Department of Mathematics University of California, Los Angeles 405 Hilgard Ave. Los Angeles, CA 90095 Abstract The goal of this work is to introduce and study fuzzy
More informationMath 117: Topology of the Real Numbers
Math 117: Topology of the Real Numbers John Douglas Moore November 10, 2008 The goal of these notes is to highlight the most important topics presented in Chapter 3 of the text [1] and to provide a few
More informationLecture 7. We can regard (p(i, j)) as defining a (maybe infinite) matrix P. Then a basic fact is
MARKOV CHAINS What I will talk about in class is pretty close to Durrett Chapter 5 sections 1-5. We stick to the countable state case, except where otherwise mentioned. Lecture 7. We can regard (p(i, j))
More informationarxiv: v1 [math.pr] 26 Mar 2008
arxiv:0803.3679v1 [math.pr] 26 Mar 2008 The game-theoretic martingales behind the zero-one laws Akimichi Takemura 1 takemura@stat.t.u-tokyo.ac.jp, http://www.e.u-tokyo.ac.jp/ takemura Vladimir Vovk 2 vovk@cs.rhul.ac.uk,
More informationDISCRETE STOCHASTIC PROCESSES Draft of 2nd Edition
DISCRETE STOCHASTIC PROCESSES Draft of 2nd Edition R. G. Gallager January 31, 2011 i ii Preface These notes are a draft of a major rewrite of a text [9] of the same name. The notes and the text are outgrowths
More informationPRISM: Probabilistic Model Checking for Performance and Reliability Analysis
PRISM: Probabilistic Model Checking for Performance and Reliability Analysis Marta Kwiatkowska, Gethin Norman and David Parker Oxford University Computing Laboratory, Wolfson Building, Parks Road, Oxford,
More informationOutlines. Discrete Time Markov Chain (DTMC) Continuous Time Markov Chain (CTMC)
Markov Chains (2) Outlines Discrete Time Markov Chain (DTMC) Continuous Time Markov Chain (CTMC) 2 pj ( n) denotes the pmf of the random variable p ( n) P( X j) j We will only be concerned with homogenous
More informationTreball final de grau GRAU DE MATEMÀTIQUES Facultat de Matemàtiques Universitat de Barcelona MARKOV CHAINS
Treball final de grau GRAU DE MATEMÀTIQUES Facultat de Matemàtiques Universitat de Barcelona MARKOV CHAINS Autor: Anna Areny Satorra Director: Dr. David Márquez Carreras Realitzat a: Departament de probabilitat,
More informationConstruction of a general measure structure
Chapter 4 Construction of a general measure structure We turn to the development of general measure theory. The ingredients are a set describing the universe of points, a class of measurable subsets along
More informationBounds on Convergence of Entropy Rate Approximations in Hidden Markov Processes
Bounds on Convergence of Entropy Rate Approximations in Hidden Markov Processes By Nicholas F. Travers B.S. Haverford College 2005 DISSERTATION Submitted in partial satisfaction of the requirements for
More information