Jordan Canonical Form

RAJEEV SINGH December 4th, 2006

In this article I shall tell the algorithm to nd the Jordan Canonical Form of a given square matrix. Then I shall give an example to illustrate the method. I shall not give any formal mathematical proof for any of the claims which I make but the proofs should be looked up from any textbook on matrix analysis.

Algorithm
Find all the eigenvalues of the given n n square matrix A. Let the eigenvalues be denoted by i , where m denotes the degeneracy of the ith eigenvalue. Find all the independent vectors satisfying the equation A i
(m) (m)

I x=0

where I is n n identity matrix and x is n-dimensional column vector. All the vectors satisfying the above equations are called rank 1 eigenvectors of A. If we are able to nd n independent rank-1 eigenvectors of A then we would be able to diagonalize it. We form a n n matrix (say R) with the independent eigenvectors by placing them as columns. Then we nd R1 . The matrix RAR1 gives the diagonal form of A. If we get lesser number of independent vectors, then we need to nd the independent vectors satisfying the equation A i
(m) 2

x=0

The vectors satisfying this equation and not satisfying the previous equation are called rank 2 eigenvectors (m) of A. We then form chains from rank-2 eigenvetors by operating them by A i I . Then we see if we have found n independent vectors including all the rank-1 and rank-2 chain of eigenvectors. Again if we have found n independent vectors, we form the matrix R and nd the product RAR1 , which gives the Jordan Canonical Form of A. Again if we get lesser number of independent vectors, we nd the independent vectors satisfying the equation A i
(m) 3

x=0

The vectors satisfying this equation and not satisfying the earlier two equations are called rank 3 eigen(m) vectors of A. Again we nd chains by successively operating by A i I , and if we succeed in nding n independent vectors we form the matrix R and nd the Jordan Canonical Form using it. We repeat the above steps to nd rank-4, rank-5, etc. eigenvectors of A till we nd n independent of them. Then we form the matrix R and use it to nd the Jordan Canonical Form of A. Next we give an illustrative example giving each step of calculation which would explain the method just outlined.

Example
Let the given square matrix be 10 24 18 54 12 1 6 3 6 1 1 8 0 16 4 1 0 2 2 0 1 4 0 6 2

1 A= 2

All the eigenvalues of this matrix are 2, i.e. it has a ve-fold degeneracy in its spectrum. 6 24 18 54 12 1 2 3 6 1 1 1 8 4 16 4 A 2I = 2 1 0 2 2 0 1 4 0 6 2 Rank-1 eigenvectors are given by (A 2I ) 1 2 6 1 1 1 1 x1 x2 x3 x4 x5 =0 =0

24 18 54 12 x1 x2 2 3 6 1 8 4 16 4 x3 0 2 2 0 x4 4 0 6 2 x5

Using Gauss elimination 6 24 18 1 2 3 1 8 4 1 0 2 1 4 0

method to solve these equations, 54 12 1 4 3 9 2 1 4 3 9 1 0 2 6 1 2 3 6 1 0 3 1 8 4 16 4 0 4 1 16 4 7 1 0 4 1 2 0 0 2 2 0 7 6 2 1 4 0 6 2 0 8 3 15 1 4 3 0 2 0 0 0 1 0 0 1 0 0 3 9 2 1 0 3 1 1 0 0 0 1 0 3 0 0 4 3 9 2 2 0 3 1 0 1 1 0 0 0 0 0 0 0 0 0

2 1 2 2 4

Thus we get three equations,

x3 + x4 = 0 x3 = x4 2x2 + 3x4 x5 = 0 x5 = 2x2 + 3x4 x1 4x2 3x3 + 9x4 2x5 = 0 x1 4x2 3x4 + 9x4 2 (2x2 + 3x4 ) = 0 x1 = 0

 Thus we have two independent rank-1 eigenvectors which are

Since we have just two independent eigenvectors of rank-1, we which are given by x1 x2 2 (A 2I ) x3 = 0 x4 x5 Now,

0 0 0 1 0 and 1 . 0 1 2 3 now need to look for rank-2 eigenvectors,

(A 2I )

1 4 1 4

6 24 18 54 12 6 24 18 54 12 1 2 3 6 1 2 3 6 1 1 1 1 8 4 16 4 8 4 16 4 1 0 2 2 0 1 0 2 2 0 1 4 0 6 2 1 4 0 6 2 0 0 0 0 0 0 0 0 0 0 2 8 6 18 4 2 8 6 18 4 6 24 18 54 12

It is easy to see that here we have just one independent equation, which is x1 + 4x2 + 3x3 9x4 + 2x5 = 0 x1 = 4x2 + 3x3 9x4 + 2x5 2 9 3 4 0 1 0 0 Thus four independent rank-2 eigenvectors are 0 , 1 , 0 and 0 0 0 0 1 1 0 0 0 The corresponding chains are given by successively operating by (A 2I ), and we get 4 0 1 1 0 2 0 0 2 0 4 3 0 0 0 1 1 1 0 2 0 1 0 3 9 0 0 3 0 1 7 0 2 1 7 0 15 3

2 0 0 0 1

1 2

0 1 2 2 4

One can check that out of all the generalized eigenvectors we have found so far only four are independent, so we need to nd rank-3 eigenvectors, which are given by x1 x2 3 (A 2I ) x3 = 0 x4 x5 Now, 0 0 2 2 6 0 0 8 8 24 0 0 0 0 0 0 6 18 4 6 18 4 18 54 12 6 24 18 54 12 1 2 3 6 1 1 8 4 16 4 1 0 2 2 0 1 4 0 6 2 0 0 0 0 1

(A 2I )

1 8 0

1 0 0 0 0 1 0 0 Hence the rank-3 eigenvectors can be taken as 0 , 0 , 1 , 0 and 0 0 0 1 0 0 0 0 The corresponding chains are given by successively operating by (A 2I ), and we get 6 0 1 1 0 0 0 1 1 1 1 0 2 2 0 1 1 1 3 0 0 12 0 1 0 1 0 4 2 0 0 2 0 0 2 6 18 0 0 0 0 3 1 1 4 1 3 0 2 2 3 0 2 0 9 0 0 0 27 0 3 0 0 8 1 9 0 2 1 1 9 27 0 3

0 0 0 0 1

1 2

12 1 4 0 2

0 0 1 1 3

Again one can check that it is not possible to nd ve independent eigenvectors with a rank-3 chain and any two rank-1 eigenvectors. Hence one has to look for a rank-3 and a rank-2 chain. There are many possible combinations of independent eigenvectors, out of which we choose chain of of one and the rst rst rank-3 the 0 4 6 0 1 1 0 1 1 1 0 1 rank-2, and the ve independent eigenvectors are 0 , 2 1 , 2 1 , 0 and 2 . 2 1 1 0 0 4 0 1 3 0 We form the following matrix using these 2 6 0 8 0 0 1 0 2 2 1 1 1 0 4 R= 0 2 0 1 1 0 4 0 1 3 0 8 Its inverse is R1 = 2 6 0 0 1 0 0 1 1 0 1 1 0 1 3 8 2 0 0 0 0 2 4 4 8

1 2

The Jordan Canonical Form is given by 2 1 0 0 0 0 2 1 0 0 0 0 2 0 0 0 0 0 2 1 0 0 0 0 2

RAR