Document

advertisement
Day 3 Markov Chains
For some interesting demonstrations of this topic visit:
http://ocw.mit.edu/OcwWeb/Mathematics/18-06Spring2005/Tools/index.htm
• Difference equation is an equation
involving differences. We can see difference
equation from at least three points of views:
as sequence of number, discrete dynamical
system and iterated function. It is the same
thing but we look at different angle.
Equations of the form:
are called discrete equations because they only model the
system at whole number time increments
Difference Equations vs Differential Equations
Dynamical system come with many different names.
Our particular interesting dynamical system is for
the system whose state depends on the input
history. In discrete time system, we call such
system difference equation (equivalent to
differential equation in continuous time).
Markov Matrices
Consider the matrix
A=
[
.1
.2
.7
]
.01 .3
.99 .3
0
.4
Properties of Markov Matrices
All entries are ≥ 0
All Columns add up to one
Note: the powers of the matrix will maintain
these properties
Each column is representing probabilities
Markov Matrices
[ ]
.1 .01 .3
A=
.2 .99 .3
.7 0
.4
1 is a eigenvalue of all Markov Matrices
Why?
Subtract 1 down each entry in the diagonal.
Each column will then add to zero which means that
the rows are dependent.
Which means that the matrix is singular
Markov Matrices
[ ]
.1 .01 .3
A=
.2 .99 .3
.7 0
.4
One eigenvalue is one all other eigenvalues have an
absolute value ≤ 1
We are interested in raising A to some powers
If 1 is an eigenvector and all other vectors are less than
1 then the steady state is the eigenvector
Note: this requires n independent vectors
Short cuts for finding eigenvectors
[
]
-.9 .01 .3
A-I =
.2 -.01 .3
det(A -1I)
.7 0
-.6
To find the eigenvector that corresponds to λ=1
Use < .6, ?? , .7> to get the last row to be zero.
Then use the top row to get the missing middle value.
<.6,33,.7>
Applications of Markov Matrices
Markov Matrices are used to when the
probability of an event depends on its
current state.
For this model, the probability of an event
must remain constant over time.
The total population is not changing over time
Markov matrices have applications in
Electrical engineering
Applications of Markov Matrices
uk+1 = Auk
Suppose we have two cities Suzhou (S) an Hangzhou
(H) with initial condition at k=0, S = 0 and H = 1000
We would like to describe movement in population
between these two cities.
Population of S and H
at time t
us+1 =
.2
uS
.9
uH+1
.1 .8
uH
[ ][
][ ]
Population of Suzhou and Hongzhou at time t+1
Column 1: .9 of the people in S stay there and .1 move to H
Column 2: .8 of the people in H stay there are and .2 move to S
Applications of Markov Matrices
uk+1 = Auk
.
us+1 =
.9
uH +1
.1
[ ][
][]
.2
.8
uS
uH
Find the eigenvalues and eigenvectors
Applications of Markov Matrices
uk+1 = Auk
.
us +1 =
.9
uH +1
.1
[ ][
][]
.2
.8
uS
uH
eigenvalues 1 and .7 (from properties of Markov
Matrices and the trace)
Eigenvectors Ker (A-I), Ker (A-.7I)
A-I -.1 .2 Ker=<2,1> A-.7I = .2 .2 Ker=<1,-1>
.1 -.2
.1 .1
[ ]
[ ]
Applications
uk+1 = Auk
us+1 =
.2
uS
.9
uH+1
.1 .8
uH
eigenvalue 1 eigenvector <2,1>
eigenvalue .7 eigenvector <-1,1>
This tells us about time and ∞
λ=1 will be a steady state, λ=.7 will disappear as t→∞
The eigenvector tells us that we need a ratio of 2:1
The total population is still 1000 so the final population
will be 1000 (2/3) and 1000 (1/3)
[ ][
][ ]
Applications
[ ][
us+1
uH +1
.9
=
.1
Initial condition at k=0, S = 0 and H = 1000
][ ]
.2
.8
uS
uH
To find the amounts after a finite number of steps
Aku
[]
k
[]
[]
2 + c2 (.7) k -1
1
1
Use the initial condition to solve for constants
0
= c1 2 + c2
-1
c1 =1000/3
1000
1
1
c2 = 2000/3
0 = c1(1)
[ ] []
Steady state for Markov Matrices
Every Markov chain will be a steady state.
The steady state will be the eigenvector for
the eigenvalue λ=1
Applications of Markov matrices
Airlines - Markov matrices are used in creating
networks for airlines to determine routes of planes.
The sum of the probabilities is 1 in each column
because all planes in a given location go somewhere
(includes possibly not moving)
Airlines want to create flight plans so they do not end
up with too many planes in one part of the world
and not enough in another.
More applications of Markov
Game theory – looking setting house rules for
casinos ensuring casinos come out ahead
Economics- Economic mobility over
generations
• http://www.facstaff.bucknell.edu/ap030/Mat
h345LAApplications/MarkovProcesses.htm
l
Homework (diff 3):
review worksheet 8.1 3-6,8,9,13
eigenvalue review worksheet 1-5
"Genius is one per cent inspiration, ninety-nine per
cent perspiration.“ Thomas Alva Edison
For More information visit:
http://people.revoledu.com/kardi/tutorial/Diffe
renceEquation/WhatIsDifferenceEquation.htm
http://www.math.duke.edu/education/ccp/mat
erials/linalg/diffeqs/diffeq2.html
Fibonacci via matrices
http://www.maths.leeds.ac.uk/applied/0380/fibonacci03.pdf
Download