STEP 3
Part A
Given a square matrix Mnxm, when doing the operation Mk without going
through the successive product of the matrix, we can operate with the
vectors and their associated values of this matrix, diagonalizing.
But we have to ask if, through a base change, we can express the matrix M
as M = S-1 AS, where A is diagonal or diagonal by blocks. These
eigenvectors have the property of, when applied to the matrix M, they
result in a linear combination of themselves (that is, a vector proportional
to them), where the value associated to each vector will play the
proportionality paper:
Operating we obtain:
This system will be compatible indeterminate, since the system (M - λnI)
will have infinite solutions. Therefore, the det (M - λnI) = 0, and thus we
obtain the so-called characteristic polynomial of M. The solutions of this
polynomial will be the different values, which will be components of the
diagonal matrix A. From the values, we substitute them in the system and
we will obtain the vectors.
The multiplicity of the roots of the characteristic polynomial will define
the algebraic multiplicity of each value (corresponding to each root). For A
to be pure diagonal matrix, it must be satisfied that the algebraic
multiplicity of each value corresponds to its geometric multiplicity (which
is calculated as the dimension of the nucleus , that, in
conclusion, will indicate the number of corresponding independent vectors
to each own value).
Then, if M is diagonalized, the result of Mk will be:
(k times)
Where Ak translates into raising to k each component of the diagonal from A.
If M is not diagonalizable, that is, if the geometric multiplicity of one of
the vectors is less than the algebraic multiplicity of this vector, then we
have to search for independent vectors to complete the base of vectors, and
these will have to comply that diagonalize for blocks M. In this case, the
vectors whose associated values have geometric multiplicity lower than
the algebraic multiplicity, these values will meet the following
requirements:
●
●
That is to say, when applying the own vector whose value has smaller
geometric multiplicity to the algebraic, it is not annulled. On the other
hand, it is canceled when applied to (M - λnI) P.
To find the vectors, we will expand the matrix M with an own vector found
(doing ), and the result we get after pivoting will be the
equations necessary to obtain the second vector (necessary to lay the base
S). If we still need a third vector, we would have to re-expand the matrix
M with the second vector found, and so on until we have the number of
vectors needed to form S.
The diagonal matrix A blocks will be as follows:
Where (In other words, has algebraic multiplicity 2).
As we can see, some 'ones' (number 1) will appear on the upper diagonal,
between the columns corresponding to the values of which we have not
found own vectors applying , but we have found vectors
making enlargements, previously described.
Part B
In order to obtain the vectors and their respective values, we will use the
MatLab, a very important tool in these calculations, because, due to the
dimensions of our stochastic matrix, performing such calculations by hand
would be very long and costly to do.
We introduce our matrix into the program and calculate the vectors and
values.
We will use the command [V, D] = eig (A). This will return V (matrix of
ectors) and D (value matrix):
[V,D] = eig (A)
>> V=
>> D=
As we can see, we have obtained 5 own vectors, along with their 5
respective values:
Part C
From obtaining the vectors and eigenvalues of our stochastic matrix, it is
evident that the calculations of the form M = S -1 AS will be simplified.
Now, we calculate the probabilities of the different situations of the reserve
in the week k = 10.
For these calculations, we will explain the form M = S -1 AS:
;
Now, we make the power k = 10 of M:
Once we obtain the result of calculating the power k = 10 of M, we will
apply the initial state we considered in Step 2. Let us remember:
We realize that we have some negative odds:
To know how to interpret this, and, taking into account from the beginning
that the probabilities have to add 1, we will divide each component of the
vector by the sum of all, we 'normalize it:
This is the final result.
