Embedded Markov Chain Limit Probabilities Limit Distribution of BD Process Chapter 5: Continuous Time Markov Chain Part 3 Kurnia Susvitasari ksusvita@sci.ui.ac.id Department of Mathematics University of Indonesia Kurnia Susvitasari SCST602204 1 / 22 Embedded Markov Chain Limit Probabilities Limit Distribution of BD Process Outlines 2 Limit Probabilities 1 Embedded Markov Chain Jump Process Class Properties of a CTMC Kurnia Susvitasari The Existence Conditions Balanced Equations 3 Limit Distribution of BD Process SCST602204 2 / 22 Embedded Markov Chain Limit Probabilities Limit Distribution of BD Process Jump Process Definition 5.6 Given a CTMC {Y (t) : t ≥ 0} defined in a state space S. The jump process is a discrete-time process {Xn = Y (Tn ) : n ∈ Z+ } where Tn is the sojourn time between the (n − 1)-th and n-th events. Kurnia Susvitasari SCST602204 3 / 22 Embedded Markov Chain Limit Probabilities Limit Distribution of BD Process Interpretation of Def’n 5.6 Jump process is derived from a CTMC by sampling this process at the jump times. Sometimes, jump process is also called embedded chain of a CTMC. The corresponding jump transition probabilities are, for i, j ∈ S Pij = Pr{Xn+1 = j|Xn = i} = Pr{Y (Tn+1 ) = j|Y (Tn ) = i} The jump process is a homogeneous Markov chain with countable state space. Kurnia Susvitasari SCST602204 4 / 22 Embedded Markov Chain Limit Probabilities Limit Distribution of BD Process Example: A Poisson Process Given a Poisson process {N(t) : t ≥ 0} with rate λ for all state i ∈ {0, 1, . . .}. We know that the inter-arrival time is exponentially distributed and independent. The process can only jump from i to i + 1; we call it pure birth process. The embedded Markov chain has the transition probabilities P0,1 = Pr{N(T1 ) = 1, N(0) = 0} Pi,i+1 = Pr{N(Ti+1 ) = i + 1, N(Ti ) = i} Kurnia Susvitasari SCST602204 5 / 22 Embedded Markov Chain Limit Probabilities Limit Distribution of BD Process Recurrent and Transient States of a CTMC Given the embedded DTMC {Xn } associated with any CTMC. Then, The transition probabilities of {Xn } form the P-matrix of CTMC. For any state i ∈ S, Pii = 0, i.e. states do not transition to themselves. Periodicity is not possible in CTMC. Kurnia Susvitasari SCST602204 6 / 22 Embedded Markov Chain Limit Probabilities Limit Distribution of BD Process Class Properties of CTMC Let i, j ∈ S, where i ̸= j. Denote “↔” be communication property. Lemma 5.2 States i ↔ j in a CTMC if i ↔ j in the embedded MC. Lemma 5.3 Since communication of states partitions embedded MC into classes, it also partitions the associated CTMC. Kurnia Susvitasari SCST602204 7 / 22 Embedded Markov Chain Limit Probabilities Limit Distribution of BD Process Class Properties of CTMC (Con’d) Lemma 5.4 A CTMC is irreducible if the embedded MC is irreducible. Lemma 5.5 State i is recurrent in a CTMC if i is recurrent in the embedded MC. Lemma 5.6 State i is transient in a CTMC if i is transient in the embedded MC. Kurnia Susvitasari SCST602204 8 / 22 Embedded Markov Chain Limit Probabilities Limit Distribution of BD Process Class Properties of CTMC (Con’d) Lemma 5.7 State i is positive recurrent in a CTMC if i is positive recurrent in the embedded MC. Lemma 5.8 Transient and recurrent are class properties of a CTMC. Kurnia Susvitasari SCST602204 9 / 22 Embedded Markov Chain Limit Probabilities Limit Distribution of BD Process Limit Probabilities Definition 5.7 Given a CTMC {X (t) : t ≥ 0} with TPF P(t). Let πj = lim Pij (t), t→∞ i.e. the MC will be in state j at t no matter where it initially comes from. If such a limit exists, we call it limit probabilities of {X (t)}. Sometimes, it is also called long-run proportion of time that the process is in state j. Kurnia Susvitasari SCST602204 10 / 22 Embedded Markov Chain Limit Probabilities Limit Distribution of BD Process Conditions for Existence Theorem 5.5 If a CTMC {X (t) : t ≥ 0} defined in state space S is (a) irreducible, and (b) positive recurrent MC, then, for all j ∈ S, πj = lim Pij (t) t→∞ exists and independent of initial state i. If limit distribution exists, we call such process ergodic. Kurnia Susvitasari SCST602204 11 / 22 Embedded Markov Chain Limit Probabilities Limit Distribution of BD Process Implication of Thm 5.5 Suppose {X (t) : t ≥ 0} has transition rates qij for all k , j ∈ S. The steady state (or limit) probabilities πj ≥ 0 are the unique solution of the system of linear equations rate at which the process enters j rate at which the process leaves j z}|{ νj πj = zX}| { qkj πk (5.1) k ̸=j X πk = 1 k ∈S P We can show that πj = νj /( i∈S νi ). We call eq. (5.1) a balance equation. Kurnia Susvitasari SCST602204 12 / 22 Embedded Markov Chain Limit Probabilities Limit Distribution of BD Process Kolmogorov’s Equation to Limit Probabilities We will show the algebraic relations of (5.1) obtained from Kolmogorov’s forward equation. The Kolmogorov’s eq. shows the instantaneous change in probabilities P(t) in an infinitesimal time h. When the process is steady (stationary), the change is zero, i.e. if the limit distribution exists =πj =πk z }| { X z }| { d Pij (t) = − lim Pij (t) νj + lim Pik (t) qkj lim t→∞ t→∞ t→∞ dt k ̸=j X ⇔ 0 = −πj νj + πk qkj . k ̸=j That is, we obtained our desired equation. Kurnia Susvitasari SCST602204 13 / 22 Embedded Markov Chain Limit Probabilities Limit Distribution of BD Process Balanced Equation in Matrix Form Given that the limit distribution exists, P ≡ lim P(t) = 1π, t→∞ where 1 is the column vector of 1’s and π is the row vector of limit probabilities πj ’s. Then, =1 z }| { πP = (π1) π = π As well, πQ = 0. Kurnia Susvitasari SCST602204 14 / 22 Embedded Markov Chain Limit Probabilities Limit Distribution of BD Process Recall a BD Process define BD process with birth rate λi and death rate µi draw a transition scheme solve balance equation by recursive show the necessary and sufficient condition for which the limit distribution exists. Kurnia Susvitasari SCST602204 15 / 22 Embedded Markov Chain Limit Probabilities Limit Distribution of BD Process Recall a BD Process Recall a BD process. When X (t) = i, let birth and death occur exponentially with rate λi and µi , respectively. The transition scheme is given as follows Kurnia Susvitasari SCST602204 16 / 22 Embedded Markov Chain Limit Probabilities Limit Distribution of BD Process Balance Eq. in BD Process The balance eq. can be expressed as follows Kurnia Susvitasari State 0 1 2 .. . Balance Eq. λ0 π0 = µ1 πi (λ1 + µ1 )π1 = µ2 π2 + λ0 π0 (λ2 + µ2 )π2 = µ3 π3 + λ1 π1 .. . n≥1 (λn + µn )πn = µn+1 πn+1 + λn−1 πn−1 SCST602204 17 / 22 Embedded Markov Chain Limit Probabilities Limit Distribution of BD Process Solving Balance Equation Using recursive method, we can solve the balance equation in terms of π0 as follows λ0 π0 µ1 λ0 λ1 λ0 π2 = π1 = π0 µ1 µ2 µ1 .. . λn−1 . . . λ0 λn−1 πn = πn−1 = π0 µn µn . . . µ1 π1 = Kurnia Susvitasari SCST602204 18 / 22 Embedded Markov Chain Limit Probabilities Limit Distribution of BD Process Solving Balance Eq. (Cont’d) Since P n πn = 1, we obtain 1 = π0 + π0 X λn−1 . . . λ0 n π0 = µn . . . µ1 1 P λn−1 . . . λ0 1+ n µn . . . µ1 Thereby, for n ≥ 1 πn = µn . . . µ1 Kurnia Susvitasari λn−1 . . . λ0 ! X λn−1 . . . λ0 1+ µn . . . µ1 n SCST602204 19 / 22 Embedded Markov Chain Limit Probabilities Limit Distribution of BD Process Solving Balance Eq. (Cont’d) The necessary condition for πn to exist is such that X λn−1 . . . λ0 n µn . . . µ1 <∞ The condition above can be shown to be sufficient. Kurnia Susvitasari SCST602204 20 / 22 Embedded Markov Chain Limit Probabilities Limit Distribution of BD Process Some Notes From the transition scheme, the chain is irreducible (every state communicates to each other). We can argue that the embedded MC is positive recurrent chain, hence the limit probabilities exist. Fact, it is very difficult to show a chain with infinitely many states to be positive recurrent. In practice, we solve the balance eq. and then determine the condition in which these probabilities exist. Kurnia Susvitasari SCST602204 21 / 22 Embedded Markov Chain Limit Probabilities Limit Distribution of BD Process “If you have knowledge, let others light their candles at it.” – Margaret Fuller Best wishes on your study! Kurnia Susvitasari SCST602204 22 / 22