Math 166: Topics in Contemporary Mathematics II Xin Ma Texas A&M University November 26, 2017 Xin Ma (TAMU) Math 166 November 26, 2017 1 / 14
A Review A Markov process is a finite sequence of experiments in which 1. Each experiment has the same possible outcomes. 2. The probabilities of the outcomes depend only on the preceding experiment. We can apply matrix multiplication to find the probability distribution X n at the n-th stage T n X 0 = X n where T is the transition matrix, X 0 is the initial distribution. We say that a matrix T is a transition (or stochastic) matrix if 1. the matrix T is square 2. all elements in the matrix T are between 0 and 1 3. the sum of all the elements in any column is 1. Xin Ma (TAMU) Math 166 November 26, 2017 2 / 14
A Review Sometimes, X n X L as n becomes large. We call X L the steady-state distribution or the limiting distribution. If a Markov process has a steady-state distribution, we call this Markov process is regular. A Markov process is regular if, and only if, the associated transition (stochastic) matrix is regular. Note: If a Markov process has a steady-state distribution, it will ALWAYS approach this column matrix X L, no matter what the initial distribution is. A transition (or stochastic) matrix T is regular if some power of T has all positive entries, in other words, all the entries are strictly greater than 0. If there is a matrix L such that T n L as n becomes large and unbounded, we call L the limiting matrix. If T is regular, the limits L exists. If T is absorbing (will be talked soon), the limits L also exists. Xin Ma (TAMU) Math 166 November 26, 2017 3 / 14
Definition: In a Markov Processes, a state is called absorbing if it has the property that when entered it is impossible to leave. For example, consider a 3-state Markov process with the transition matrix T. If the second state is absorbing, then for any natural number n we always have 0 0 T n 1 = 1 0 0 Definition: A Markov process is called absorbing if 1. There is at least one absorbing state. 2. It is possible to move from any nonabsorbing state to one of the absorbing states in a finite number states. The transition matrix for an absorbing Markov process is said to be an absorbing stochastic matrix. Xin Ma (TAMU) Math 166 November 26, 2017 4 / 14
Example: A mouse house has 3 rooms: Room A, B, and C. Room A has cheese in it. Every minute, the mouse makes a choice. If the mouse is in Room B, the probability is 0.6 that it will move to Room A and 0.4 that it will move to Room C. If the mouse is in Room C, the probability that it will stay there is 0.1, the probability that it will move to Room A is 0.7, and the probability that it will move to Room B is 0.2. Once the mouse enters Room A, it never leaves. a. Write the transition matrix T for this Markov process. Is it an absorbing Markov process? Xin Ma (TAMU) Math 166 November 26, 2017 5 / 14
Example: A mouse house has 3 rooms: Room A, B, and C. Room A has cheese in it. Every minute, the mouse makes a choice. If the mouse is in Room B, the probability is 0.6 that it will move to Room A and 0.4 that it will move to Room C. If the mouse is in Room C, the probability that it will stay there is 0.1, the probability that it will move to Room A is 0.7, and the probability that it will move to Room B is 0.2. Once the mouse enters Room A, it never leaves. a. Write the transition matrix T for this Markov process. Is it an absorbing Markov process? A B C A 1 0.6 0.7 T = B 0 0 0.2 ; Yes, Because A is absorbing and C 0 0.4 0.1 P(B A) = 0.6 > 0 and P(C A) = 0.7 > 0 Xin Ma (TAMU) Math 166 November 26, 2017 5 / 14
Remark: If a Markov process is absorbing, then there is at least a ONE on the main diagonal of the associated transition matrix. This is a necessary condition but not sufficient. You also need to verify condition 2 in the definition. b. If there are a bunch of mice in the house, with 10% in Room A, 60% in Room B, and 30% in Room C, what is the room distribution after 5 minutes? Xin Ma (TAMU) Math 166 November 26, 2017 6 / 14
Remark: If a Markov process is absorbing, then there is at least a ONE on the main diagonal of the associated transition matrix. This is a necessary condition but not sufficient. You also need to verify condition 2 in the definition. b. If there are a bunch of mice in the house, with 10% in Room A, 60% in Room B, and 30% in Room C, what is the room distribution after 5 minutes? 0.1 0.995839 X 0 = 0.6. Then X 5 = T 5 X 0 = 0.00135 0.3 0.002811 Xin Ma (TAMU) Math 166 November 26, 2017 6 / 14
c. If the mouse is initially in Room C, what are the probabilities of being in each room after 5 minutes? Xin Ma (TAMU) Math 166 November 26, 2017 7 / 14
c. If the mouse is initially in Room C, what are the probabilities of being in each room after 5 minutes? 0 0.99597 Z 0 = 0. Then Z 5 = T 5 Z 0 = 0.00178 1 0.00225 d. Find T 5 and guess what the limiting matrix L should be. Xin Ma (TAMU) Math 166 November 26, 2017 7 / 14
c. If the mouse is initially in Room C, what are the probabilities of being in each room after 5 minutes? 0 0.99597 Z 0 = 0. Then Z 5 = T 5 Z 0 = 0.00178 1 0.00225 d. Find T 5 and guess what the limiting matrix L should be. 1 0.99508 0.99597 1 1 1 T 5 = 0 0.00136 0.00178 We may expect that L = 0 0 0. 0 0.00356 0.00225 0 0 0 e. Find the long-term behavior if the mouse starts at B and C. Xin Ma (TAMU) Math 166 November 26, 2017 7 / 14
c. If the mouse is initially in Room C, what are the probabilities of being in each room after 5 minutes? 0 0.99597 Z 0 = 0. Then Z 5 = T 5 Z 0 = 0.00178 1 0.00225 d. Find T 5 and guess what the limiting matrix L should be. 1 0.99508 0.99597 1 1 1 T 5 = 0 0.00136 0.00178 We may expect that L = 0 0 0. 0 0.00356 0.00225 0 0 0 e. Find the long-term behavior if the mouse starts at B and C. 0 0 1 Y 0 = 1 Z 0 = 0. The long-term behavior: LY 0 = 0 and 0 1 0 1 LZ 0 = 0 as well. 0 Xin Ma (TAMU) Math 166 November 26, 2017 7 / 14
Remark: If a absorbing Markov process has only one absorbing state, say the ith state, then the long term probability of ending in the ith state is one no matter what the initial distribution is. This is why we call it absorbing. And all the entries of ith row of the limiting matrix are 1 and the other entries are 0 as the example above shows. Now let us consider an example of a Markov process with more than one absorbing states. Example: A mouse house has 3 rooms: Room A, B, and C. Room A has cheese in it and Room B has a trap in it. Every minute, the mouse makes a choice. If the mouse is in Room C, the probability is 0.2 that it will stay there, 0.3 that it will move to Room A, and 0.5 that it will move to Room B. Once the mouse enters either Room A or Room B, it never leaves. Xin Ma (TAMU) Math 166 November 26, 2017 8 / 14
a. Write the transition matrix T for this Markov process. Is it an absorbing Markov process? Xin Ma (TAMU) Math 166 November 26, 2017 9 / 14
a. Write the transition matrix T for this Markov process. Is it an absorbing Markov process? A B C A 1 0 0.3 T = B 0 1 0.5 ; C 0 0 0.2 Yes, Because states A and B are absorbing and P(C B) = 0.5 > 0 and P(C A) = 0.3 > 0. b. If there are a bunch of mice in the house, with 10% in Room A, 60% in Room B, and 30% in Room C, what is the room distribution after 5 minutes? Xin Ma (TAMU) Math 166 November 26, 2017 9 / 14
a. Write the transition matrix T for this Markov process. Is it an absorbing Markov process? A B C A 1 0 0.3 T = B 0 1 0.5 ; C 0 0 0.2 Yes, Because states A and B are absorbing and P(C B) = 0.5 > 0 and P(C A) = 0.3 > 0. b. If there are a bunch of mice in the house, with 10% in Room A, 60% in Room B, and 30% in Room C, what is the room distribution after 5 minutes? 0.1 X 0 = 0.6. Then X 5 = T 5 X 0 = 0.3 0.212464 0.78744 9.6 10 5 Xin Ma (TAMU) Math 166 November 26, 2017 9 / 14
c. If the mouse is initially in Room C, what are the probabilities of being in each room after 5 minutes? Xin Ma (TAMU) Math 166 November 26, 2017 10 / 14
c. If the mouse is initially in Room C, what are the probabilities of being in each room after 5 minutes? 0 0.37488 X 0 = 0. Then X 5 = T 5 X 0 = 0.6248 1 3.2 10 4 d. Find T 5 and T 50 and guess what the limiting matrix L should be. Xin Ma (TAMU) Math 166 November 26, 2017 10 / 14
c. If the mouse is initially in Room C, what are the probabilities of being in each room after 5 minutes? 0 0.37488 X 0 = 0. Then X 5 = T 5 X 0 = 0.6248 1 3.2 10 4 d. Find T 5 and T 50 and guess what the limiting matrix L should be. By calculator: 1 0 0.37488 1 0 0.375 T 5 = 0 1 0.6248 and T 50 = 0 1 0.625 0 0 3.2 10 4 0 0 1.1259 10 35 1 0 0.375 Then we may expect L = 0 1 0.625. 0 0 0 Xin Ma (TAMU) Math 166 November 26, 2017 10 / 14
e. Find the long-term behavior if the mouse starts at C. Xin Ma (TAMU) Math 166 November 26, 2017 11 / 14
e. Find the long-term behavior if the mouse starts at C. 0 Z 0 = 0. The long-term behavior: 1 0.375 LZ 0 = 0.625. 0 Xin Ma (TAMU) Math 166 November 26, 2017 11 / 14
Remark: 1. If a Markov process has 2 or more absorbing states, then the long term probabilities of ending in the nonabsorbing states will be 0 and the long term probabilities of ending in the absorbing states will depend on the initial distribution. 2. We can deduce the limiting matrix L by computing T n for large n (usually n=50 is sufficiently large and then round your answers to four decimal places). Then find the ending distribution according to the initial distribution X 0, say LX 0. Note: No matter how many absorbing states a Markov process has, eventually everything will be absorbed by the absorbing states. Xin Ma (TAMU) Math 166 November 26, 2017 12 / 14
Example: Determine if the following stochastic transition matrices are absorbing. If absorbing, find the limiting matrix. 0.3 0 0.2 a. A = 0.4 1 0.7 0.3 0 0.1 Xin Ma (TAMU) Math 166 November 26, 2017 13 / 14
Example: Determine if the following stochastic transition matrices are absorbing. If absorbing, find the limiting matrix. 0.3 0 0.2.0 0 0 a. A = 0.4 1 0.7 (Yes). L = 1 1 1 since there is only one 0.3 0 0.1 0 0 0 absorbing state (also try A 50 to verify this). 0 1 0.3 b. B = 1 0 0.3 0 0 0.4 Xin Ma (TAMU) Math 166 November 26, 2017 13 / 14
Example: Determine if the following stochastic transition matrices are absorbing. If absorbing, find the limiting matrix. 0.3 0 0.2.0 0 0 a. A = 0.4 1 0.7 (Yes). L = 1 1 1 since there is only one 0.3 0 0.1 0 0 0 absorbing state (also try A 50 to verify this). 0 1 0.3 b. B = 1 0 0.3 (no) since there is no absorbing state (no entries in 0 0 0.4 diagonal is one). Xin Ma (TAMU) Math 166 November 26, 2017 13 / 14
c. 0.3 0 0.4 0.5 1 0 0.2 0 0.6 Xin Ma (TAMU) Math 166 November 26, 2017 14 / 14
c. d. 0.3 0 0.4 0.5 1 0 (Yes) second state is absorbing. and 1st 2nd and 0.2 0 0.6 3rd 1st 2nd. 1 0 0 0 0.2 0.6 0 0.8 0.4 Xin Ma (TAMU) Math 166 November 26, 2017 14 / 14
0.3 0 0.4 c. 0.5 1 0 (Yes) second state is absorbing. and 1st 2nd and 0.2 0 0.6 3rd 1st 2nd. 1 0 0 d. 0 0.2 0.6 (No). Though 1st state is absorbing. but the other 0 0.8 0.4 two states cannot go to 1st state. e. 1 0.5 0 0.4 0 0.2 0 0.5 0 0 1 0 0 0.3 0 0.1 Xin Ma (TAMU) Math 166 November 26, 2017 14 / 14
c. d. 0.3 0 0.4 0.5 1 0 (Yes) second state is absorbing. and 1st 2nd and 0.2 0 0.6 3rd 1st 2nd. 1 0 0 0 0.2 0.6 (No). Though 1st state is absorbing. but the other 0 0.8 0.4 two states cannot go to 1st state. 1 0.5 0 0.4 e. 0 0.2 0 0.5 0 0 1 0 (Yes.) The 1st and 3rd state are absorbing. No 0 0.3 0 0.1 other two states can go to the 3rd state. But both of them can go to the 1st state. Xin Ma (TAMU) Math 166 November 26, 2017 14 / 14