程序代写代做代考 C Lecture VII

Lecture VII
• The Jordan Canonical Form • Examples and Applications
Fall 2020 Prof.Jiang@ECE NYU 295

Review of Canonical Forms
 If A is an n  n matrix with distinct eigenvalues, then there exists a nonsingular matrix P s.t.
P1 AP  diag i .
 If A is Hermitian (possibly having repeated eigenvalues),
 a unitary matrix U such that U*AU diag .
i
Fall 2020 Prof.Jiang@ECE NYU 296

A Motivating Example
As stated previously, not every matrix can be transformed into a canonical diagonal form. For example,
A1 b, b0 0 1

cannot be transformed into a diagonal matrix. Fall 2020 Prof.Jiang@ECE NYU 297

Jordan Canonical Form
Theorem (Jordan):
Let A be an n  n matrix whose different eigenvalues
are  , ,  with multiplicities m ,, m : 1s 1s
    i.e.,  nonsingular P such that
s
 i1
m
det I  A 
Then, A is transformable into a Jordan canonical form.
P1AP  blockdiagi  J
Fall 2020 Prof.Jiang@ECE NYU 298
i i

Fall 2020
i
Prof.Jiang@ECE NYU 299
Theorem (Jordan), cont’d: P1AP  blockdiagi  J
where
i 0  0
1   0 i
i        00 1 0
Jordan Block
i 0 0 1 

Comments
 In some texts, J T is used as Jordan form.  Different Jordan blocks, say i ,  j may be
associated with the same eigenvalues.
s ThetotalnumberofJordanblocks:ss n.
Fall 2020 Prof.Jiang@ECE NYU 300

Fall 2020
Prof.Jiang@ECE NYU 301
Illustration via 3×3 matrices If a 33 matrix A has an eigenvalue 1
of multiplicity three, then it may be reduced into one of the following Jordan forms:
1 0 0 1 0 0 J0  0,J0  0,
1121 00 01
11 1 0 0
J1  0. 31
01 1

Remark 1
The distinct Jordan forms Ji , Jk , i  k, are not similar to each other.
Fall 2020 Prof.Jiang@ECE NYU 302

Remark 2
When each Jordan block  ( ) in the Jordan ii
form J is one-dimensional (i.e. n  1) and s  n, i
the Jordan matrix J becomes diagonal.
Fall 2020 Prof.Jiang@ECE NYU 303

Application to Matrix Analysis of Differential Equations
Given a set of 1st-order differential equations

x(t)Ax(t), x(0)n,
applying the transformation y  P1x yields: 
y(t)P1APy(t) : Jy(t).
y1 
 yi(t)yi(t),yimi,y .
Fall 2020 Prof.Jiang@ECE NYU
304
i
ys  

Comment
So, with the help of Jordan canonical form, solving differential equations can be reduced down to solving lower-order (disjoint!) differential equations.
(see a forthcoming lecture.)
Fall 2020 Prof.Jiang@ECE NYU 305

Principal Vectors
In order to develop a constructive method for P resulting in Jordan form, let’s introduce the notion of principal vector, or generalized eigenvector,
which is a generalization of eigenvector.
Fall 2020 Prof.Jiang@ECE NYU 306

Principal Vectors
A (possibly zero) vector p is a principal vector of grade g  0 belonging to the eigenvalue i if
iIAg p0,
for which g is the smallest non-negative integer.
Fall 2020 Prof.Jiang@ECE NYU 307

Examples
• The vector p = 0 is the principal vector of grade 0.
• The (nonzero) eigenvectors are the principal vectors of grade 1.
Fall 2020 Prof.Jiang@ECE NYU 308

Motivating Question
In case of transformation to diagonal canonical form, i.e., P1 AP  diag i ,
the columns of P are linearly independent eigenvectors.
What about the matrix P in Jordan form?
How to construct P from principal vectors? Fall 2020 Prof.Jiang@ECE NYU 309

gi
i
Linear Spaces
Define the linear space composed of all principal vectors of grade  g belonging to i :
Pp|IAp0 g 
i.e., the null space of i I  Ag . Clearly,
P   P   P   0i1i2i
Fall 2020 Prof.Jiang@ECE NYU 310

An Interesting Result
Let A be an n  n matrix with the distinct
eigenvalues 1, , s , 1 s  n, with
multiplicities m , , m . 1s
Then, every vector x  n can be written as x  p1  p2  ps
where pi is a uniquely defined principal vector associated with i of grade  mi .
Fall 2020 Prof.Jiang@ECE NYU 311

Comment 1
A special, but interesting, case is when there are n linearly independent eigenvectors, say,
c1, , cn. In this case, i scalars s.t. xc1cn :p1pn.
1n
Fall 2020 Prof.Jiang@ECE NYU 312

Comment 2
Its proof relies upon the well-known Cayley- Hamilton theorem; see any standard matrix or linear algebra textbook.
Fall 2020 Prof.Jiang@ECE NYU 313

1 0 0 Consider the matrix A  7 1 0.
Example

0 0 2 
 Compute its eigenvalues and the associated eigenvectors. * Can each column be written as a linear combination
of eigenvectors?
* Show that each column can be written
as a unique representation of principal vectors.
Fall 2020 Prof.Jiang@ECE NYU 314

Fall 2020
Prof.Jiang@ECE NYU 315
Answer
  1 m  2,   2 m 1. 1122
 The eigenvectors of 1 are of the form col0,1,0,  any nonzero scalar.
The eigenvectors of 2 are of the form   col 0, 0,1,  any nonzero scalar.
 P p2|p2col0,0,.
P p1|p1col,,0
21 12


Cayley-Hamilton Theorem Revisited
For any n  n matrix A,
(A)An 1An1n1AnIO
where  is the characteristic polynomial of A, i.e.,
detIAn n ini. i1
Fall 2020 Prof.Jiang@ECE NYU
316

Fall 2020
Prof.Jiang@ECE NYU
317
 () 2 23. A
Example
ConsiderA7 4.Verifythat 8 5

1) The characteristic polynomial A () is:
2)(A)A22A3I00 0.
A
0 0 

Another Proof
Define the n  n matrix of signed cofactors: C   cof I  A.
T
Then, usingMcofM (detM)I,
IACT I. In addition,
CT n1C0 Cn1 Cn
for constant matrices Ci ‘ s.
Fall 2020 Prof.Jiang@ECE NYU 318

C0 I
C  AC   I
101

Proof (cont’d)
By identification of the coefficients of equal powers of  gives
Cn1  ACn2  n1I ACn1 nI.
Multiplying the first eq. by An , the second by An1,…,
and then adding them up leads to: O  AI.
Fall 2020 Prof.Jiang@ECE NYU 319

Question
How to compute principal vectors for a given matrix?
Fall 2020 Prof.Jiang@ECE NYU 320

A Motivating Example
Consider a 22 Jordan block J  0. 1 
Denote P  x1 x2  that transforms A into J. 
Namely, P1AP  J. So, we have AP  PJ, or
Ax1 x2x1 x2 0    1  
 Ax2  x2 , so x2 is an eigenvector;
(AI)x1  x2, so x1 is a principal vector (of grade 2). Fall 2020 Prof.Jiang@ECE NYU 321


Comment
Usually, x , x is called a Jordan Basis for 1 2
this 22 matrix A. In order words,
the JCF transformation matrix P is composed
of a Jordan basis, or a set of linearly independent eigenvectors and principal vectors.
Fall 2020 Prof.Jiang@ECE NYU 322

General Procedure
Step 1: Solve the characteristic equation AIz1 0.
Step 2 : For each independent z1 , solve AIz2 z1
2 wherez2 clearlysolves AI z2 0.
Collect only those z2 which are linearly independent
with the previously found eigenvectors z1.
Fall 2020 Prof.Jiang@ECE NYU 323

General Procedure
Step 3: For each independent z 2 , solve AIz3 z2
3 wherez3 clearlysolves AI z3 0.
Collect only those z3 which are linearly independent with the previously found vectors z1, z2.
Step 4: Continue in this way till the total number of independent eigenvectors and principal vectors equals
to the (algebraic) multiplicity of .
Fall 2020 Prof.Jiang@ECE NYU 324

General Procedure
Step 4 (cont’d): Denote
1 2 mm m1 1
x,x,,xz,z ,,z and
P  x1, x2 ,, xm . 
eigenvector
Therefore,
P1 AP  J (associated with eigenvalue ).
Fall 2020 Prof.Jiang@ECE NYU 325

Comments
• Not any arbitrary choice of linearly independent principal vectors would lead to a correct transformation matrix P.
For example, at Step 2, the linearly independent principal vectors z2 are chosen according to
AIz2 z1 but NOT :
IAz2 z1.

Fall 2020 Prof.Jiang@ECE NYU 326
See (the 1960 book of Gantmacher, Vol.1, Chap. VI, Section 8) for another general method of constructing a transformation matrix.

Fall 2020
x4 :AIx3 
Prof.Jiang@ECE NYU 327
More on Jordan Basis
Without going into the full details in proving Jordan’s Theorem, let’s illustrate the concept of Jordan basis and its use in the canonical transformation.
Consider a principal vector v of grade g  n  4. Define:
x1 : v
x2 :AIx1
 x3 :AIx2
Jordan Basis

That is,
eigenvector
Jordan Basis (cont’d)
Then, the 44 matrix A can be transformed into the Jordan canonical form:
 0 0 0
100 J  010
Principal vector of grade 4
0 0 1  
P1APJ, Px1,x2,x3,x4.
Fall 2020 Prof.Jiang@ECE NYU 328

Comment
If we define P  x4,x3,x2,x1 , then A is transformed 
into the Jordan canonical form J T , i.e.:  1 0 0
Fall 2020
Prof.Jiang@ECE NYU 329
010 1T 
P APJ 0 0  1. 0 0 0 


A More Complex Case
If AI has rank n2, i.e. its null space is of dimension 2, then  two linearly independent eigenvectorsto AIq0.
Thus, we need n  2 linearly independent principal vectors. In this case, the Jordan basis takes the form
v1,v2,,vk and u1,u2,,ul , kln. 
So, A is transformed into the Jordan canonical form P APdiag J1,J2 .
1 
Fall 2020 Prof.Jiang@ECE NYU 330

Fall 2020
Prof.Jiang@ECE NYU 331
Exercise 1
Find a transformation matrix P to bring the following matrix
M1 b, b0 0 1

into the Jordan Canonical Form J1 0.
1 1 

Fall 2020
Prof.Jiang@ECE NYU 332
Exercise 2
Find a transformation matrix to bring the following matrix into a Jordan form:
1 11 1
3 3 5 4  A  8 4 3 4 
15 10 11 11 

Solution:
Fall 2020
Prof.Jiang@ECE NYU 333
0 1 0 1 1 5 0 5
P , 0 4 1 5
1110 12 
1 1 0 0
0 1 1 0 J  .
0 0 1 0 0 0 0 1


I  A becomes (after elementary operations on rows and columns: 100 0
0 1 0 0   .
0 0 1 0  
 3 0 0 0 1 
Therefore, the matrix has two elementary divisors:
 1 and  13 ,
which give two Jordan blocks, respectively:
1 1 0 J1,J0 1 1.
12 0 0 1

See (the 1960 book of Gantmacher, Vol.1, pp.160-164)
for the details.
Fall 2020 Prof.Jiang@ECE NYU 334

Practicing Problems for Midterm
1. Compute the eigenvalues of the matrix A7 2
41 
and transform it to one of the canonical forms.
Fall 2020 Prof.Jiang@ECE NYU 335

Practicing Problems for Midterm
2. Consider the block diagonal matrix A0
Fall 2020
Prof.Jiang@ECE NYU 336
A 1 ,withAnn,nnn. iii12
0A 2
Show that the eigenvalues of A are those of
A and A . 12

Practicing Problems for Midterm
3. Assume A is a nonsingular matrix. If  is an eigenvalue of A with eigenvector x,
showthat1 isaneigenvalueofA1.
In addition, give an eigenvector associated
with 1.
Fall 2020 Prof.Jiang@ECE NYU 337

Practicing Problems for Midterm
4. Show that A   0 1  cannot be transformed into 0 0

a diagonal matrix under any similarity transformation.
Fall 2020 Prof.Jiang@ECE NYU 338

Practicing Problems for Midterm
5. For any given 2  2 real orthogonal matrix U , one of the following must hold:
(i) U   cos sin  for some ; sin cos

(ii)U0 1cos sinforsome.
1 0sin cos  
(Only for those who love math proof!)
Fall 2020 Prof.Jiang@ECE NYU 339

Practicing Problems for Midterm
6. Show that JT is similar to J. That is, 01 01
  JT  J. 
10 10 
Fall 2020 Prof.Jiang@ECE NYU 340

Practicing Problems for Midterm
7. Assume that A, D are invertible matrices. Show that
 A B 1  A1 A1BD1  0D0 D1 .

Fall 2020 Prof.Jiang@ECE NYU 341

Practicing Problems for Midterm
8. Assume that A, D are invertible matrices. Show that
A B1 A1 A1BECA1 A1BE C D  ECA1 E 
 
where E is the inverse of the Schur complement 1
ofA: EDCA1B .
Note: A Very Useful Identity.
Fall 2020 Prof.Jiang@ECE NYU 342

Practicing Problems for Midterm
9. Reduce the following matrix into a canonical diagonal form:
Fall 2020
Prof.Jiang@ECE NYU 343
A M 022  0 M
 22  where
M0 1 1 0


Practicing Problems for Midterm
10. Reduce the following matrix into a Jordan canonical form:
Fall 2020
Prof.Jiang@ECE NYU 344
3 2 1 A0 3 0

0 0 3 

Practicing Problems for Midterm
11. Rank Inequalities (See Horn-Johanson text, page 13)
 Sylvester inequality Amk, Bkn, wehave
rankArankBkrankABminrankA, rankB.  Frobenius inequality
Amk, Bkp, Cpn, wehave
rankAB  rankBC  rankB + rankABC
with equality iff there are matrices X and Y such that
B  BCX YAB.
Fall 2020 Prof.Jiang@ECE NYU 345

Fall 2020
Prof.Jiang@ECE NYU 346
1. For the matrix
1 0 1
Homework #7
A0 2 0, 
0 0 1 
identify the spaces P and the principal g
vectors of grade 2.

Fall 2020
Prof.Jiang@ECE NYU 347
Homework #7
2. Express the following vectors as unique representations of principal vectors found in Problem 1:
 2 x9
0  , x9.3.


84 
0  

Fall 2020
Prof.Jiang@ECE NYU 348
Homework #7
3. Can you transform the following matrix into a Jordan form:
  
A0  , 0? 
0 0  