THE EIGENVALUE PROBLEM
Download
Report
Transcript THE EIGENVALUE PROBLEM
BY
YAN RU LIN
SCOTT HENDERSON
NIRUPAMA GOPALASWAMI
GROUP 4
11.1 EIGENVALUES &
EIGENVECTORS
Definition
๏ An eigenvector of a n x n matrix A is a nonzero vector
x such that ๐จ๐ = ฮป๐ for some scalar ฮป.
๏ A scalar ฮป is called an eigenvalue of A if there is a
nontrivial solution x of ๐จ๐ = ฮป๐; such an x is called an
eigenvector corresponding to ฮป.
D.C. Lay, "Eigenvectors and Eigenvalues," in Linear Algebra and Its Applications, 3rd ed. Boston, MA: Pearson, 2006, ch. 5, pp. 301-372
Formulation
๏ ๐จ๐ = ฮป๐
๏ This is also equivalent to (๐จ โ ฮป๐ฐ)๐ = ๐
๏ Now it can be solved easily using a determinant
Essentially, for some value of ฮป given a transformation matrix A, there
may exist a vector x such that the equation is satisfied. If ฮป and this
particular vector x do exist, then we call ฮป the eigenvalue and the x the
corresponding eigenvector.
Example
1. ๐ด๐ ๐ ๐ข๐๐
5 โ1
=๐จ
3 10
5 โ1
๐ = ฮป๐
3 10
5 โ1
3.
๐ โ ฮป๐ฐ๐ = ๐
3 10
5โฮป
โ1
4.
๐=๐
3
10 โ ฮป
5. Solve now using a determinant
2.
Importance
๏ Eigenvalues and eigenvectors find numerous
applications in these areas:
๏ Differential Equations
๏ Dynamical systems
๏ Engineering design
๏ Chemistry and physics
๏ Schrödinger equation (quantum mechanics)
๏ Vibration analysis
BRAINBITE
A. ฮป=2 B. ฮป=0
C. ฮป=-3 D. ฮป=1
Remember thatโฆ
๐จ๐ = ฮป๐
x is given along with A, so ฮป is solved easily
http://www.maths.usyd.edu.au/u/UG/JM/MATH1014/Quizzes/quiz10.html
Answer: C
11.2 EIGENVALUES SOLUTION
PROCEDURE AND
APPLICATIONS
11.2 Eigenvalues Solution Procedure
and Applications
๏ Ax = ๏ฌx ๏ (A-๏ฌI)x = 0
๏ x=0 is a trivial solution
๏ Non-trivial solutions exist if and only if:
a11 ๏ญ ๏ฌ
det(A ๏ญ ๏ฌI ) ๏ฝ
a21
๏
an1
๏
a1n
a22 ๏ญ ๏ฌ ๏
๏
๏
a2 n
๏
a12
an 2
๏ ann ๏ญ ๏ฌ
๏ฝ0
๏ Resulting algebraic equation is called the characteristic
๏
๏
๏
๏
equation.
Characteristic polynomial- nth-order polynomial in ๏ฌ
Roots are the eigenvalues {๏ฌ1, ๏ฌ2, โฆ, ๏ฌn}
Solution space is called eigenspace corresponding to {๏ฌ1,
๏ฌ2, โฆ, ๏ฌn}
The solutions obtained are called eigenvectors
Eigenvalue Example
๏ Characteristic matrix
๏ฉ1 2 ๏น
๏ฉ1 0๏น ๏ฉ1 ๏ญ ๏ฌ
A ๏ญ ๏ฌI ๏ฝ ๏ช
๏ญ ๏ฌ๏ช
๏ฝ๏ช
๏บ
๏บ
๏ซ3 ๏ญ 4๏ป
๏ซ0 1๏ป ๏ซ 3
๏ Characteristic equation
2 ๏น
๏ญ 4 ๏ญ ๏ฌ ๏บ๏ป
A ๏ญ ๏ฌI ๏ฝ (1๏ญ ๏ฌ)(๏ญ4 ๏ญ ๏ฌ) ๏ญ (2)(3) ๏ฝ ๏ฌ2 ๏ซ 3๏ฌ ๏ญ10 ๏ฝ 0
๏ Eigenvalues: ๏ฌ1 = -5, ๏ฌ2 = 2
11.2 Subsection(1) -Quick Tips
๏ An n x n matrix A means that
are n values to x, and there
will be n eigenvectors and
eigenvalues even if some are
duplicated
๏ The eigenvalues of a triangular
matrix are the entries on its
main diagonal
๏ Consider that since ฮป is scalar,
A must act on eigenvectors
only to โstretchโ x and not to
change its direction (see figure)
Unknown. (2011, Oct
27). Eigenvalues and
eigenvectors [Online].
Available: http://en.wikipedia.o
rg/wiki/Eigenvalues_and_eigen
vectors
Example
๏ Click here to view a demo on eigenvalues and
eigenvectors
๏ http://web.mit.edu/18.06/www/Demos/eigen-applet-
all/eigen_sound_all.html
11.2 Subsection(2)-Determining Eigenvectors
๏ First determine eigenvalues: {๏ฌ1, ๏ฌ2, โฆ, ๏ฌn}
๏ Then determine eigenvector corresponding to each
eigenvalue:
(A ๏ญ ๏ฌI)x ๏ฝ 0 ๏ (A ๏ญ ๏ฌk I)xk ๏ฝ 0
๏ Eigenvectors determined up to scalar multiple
๏ Distinct eigenvalues
๏ Produce linearly independent eigenvectors
๏ Repeated eigenvalues
๏ Produce linearly dependent eigenvectors
๏ If n roots are equal then the eigenvalues are said
to of multiplicity n.
Eigenvector Example
๏ Eigenvalues
๏ฉ1 2 ๏น ๏ฌ1 ๏ฝ ๏ญ5
A๏ฝ๏ช
๏บ ๏ฌ ๏ฝ2
3
๏ญ
4
๏ซ
๏ป
2
๏ Determine eigenvectors: Ax = ๏ฌx
x1 ๏ซ 2 x2 ๏ฝ ๏ฌx1
3x1 ๏ญ 4 x2 ๏ฝ ๏ฌx2
๏
๏ Eigenvector for ๏ฌ1 = -5
(1 ๏ญ ๏ฌ ) x1 ๏ซ 2 x2 ๏ฝ 0
3x1 ๏ญ (4 ๏ซ ๏ฌ ) x2 ๏ฝ 0
6 x1 ๏ซ 2 x2 ๏ฝ 0
3x1 ๏ซ x2 ๏ฝ 0
๏ Eigenvector for ๏ฌ1 = 2
๏ญ x1 ๏ซ 2 x2 ๏ฝ 0
3x1 ๏ญ 6 x2 ๏ฝ 0
๏ฉ1๏น
x1 ๏ฝ ๏ช ๏บ
๏ซ๏ญ 3๏ป
๏ฉ2๏น
x2 ๏ฝ ๏ช ๏บ
๏ซ1๏ป
BRAINBITE
http://www.maths.usyd.edu.au/u/UG/JM/MATH1014/Quizzes/quiz12.html
Answer : c
11.2.2 APPLICATIONS TO ELEMENTARY
SINGULARITIES IN THE PHASE PLANE
๏ Consider a linear system of ODEs given by
๏ If the eigenvalues ฮป is real
Criteria
Type
ฮป<0
Stable node
ฮป>0
Unstable node
ฮป > 0 and ฮป < 0
Saddle
๏ If eigenvalues are complex of the form ๐ + ๐๐
Criteria
Type
a<0
Stable focus
a=0
Centre
a>0
Unstable focus
11.2. subsection(3)Special matrices in exercises
(1) Markov Matrix
๏ Let
A=
๏ The sum of elements of row or column sum to unity.
๏ One of the eigenvalue of Markov matrix is 1.
๏ The rows of [A-I]sum to zero
๏ [A-I] is singular and columns of A-I are linearly dependent.
M.D. Greenberg, "The Eigenvalue Problem," in Advanced Engineering Mathematics,
2nd ed. Upper Saddle River, New Jersey: Prentice Hall, 1998, ch. 11.3
(2)Tridiagnol matrix
๏ A Tridiagnol matrix is one in which all element are
zero except the principal diagonals and its two
adjacent diagonals .
๏ Eigenvalues are given by
(3) Generalized eigenvalue problem
๏ If Bโ 1 then Ax=ฮปBx is called generalized eigenvalue
problem.
๏ Characteristic equation got by det(A - ฮปB)x=0
๏ Eigenvectors given by (A - ฮปB)x=0
(4) Cayley hamilton theorem
๏ Theorem- The characteristic equation of any square
matrix A is
ฮปn+ ฮฑ1 ฮปn-1 +โฆ. ฮฑn ฮป =0
then
An+ ฮฑ1 An-1+โฆ+ ฮฑn -1A+ ฮฑn I=0.
i.e A satisfies characteristic equation.
BRAINBITE
http://www.maths.usyd.edu.au/u/UG/JM/MATH1014/Quizzes/quiz12.html
Answer : a
11.3 SYMMETRIC MATRICES
๏ A square matrix is symmetric if A = AT. This means
that each element aij = aji, as figure[1].
๏ A symmetric matrix needs not have
real numbers as elements. However,
when it does, it has the remarkable
property of having only real
eigenvalues.
๏ Proof : recall, a complex number ๐ง =๐ + ๐๐, then, the
conjugate of z is defined to be ๐ง = ๐ โ ๐๐, and ๐ด๐ต =
(๐ด)(๐ต), ๐ง๐ง = ๐ง 2 โฆ[2]
[1] http://www.aiaccess.net/English/Glossaries/GlosMod/e_gm_symmetric_matrix.htm
[2]Peter V. OโNeil, โEigenvalues,Diagnolization and Special Matricesโ in Advanced Engineering
Mathematics 5th edition. Birmingham, AL: B. Stenquist, 2003, ch. 8.3, pp. 354-362.
Let A be an n x n matrix. Let ๐ธ be the eigenvector
corresponding to its eigenvalue ๐, and get
๐ธ=
๐1
๐2
โฎ
๐๐
, ๐ธ ๐ = ๐1 , ๐2 โฏ ๐๐ ,
๐ธ ๐ ๐ด๐ธ = ๐ธ ๐ ๐๐ธ = ๐๐ธ ๐ ๐ธ = ๐ ๐1 , ๐2 โฏ ๐๐
= ๐( ๐1
2
+ ๐2
2
๐1
๐2
โฎ
๐๐
+ โฏ + ๐๐ 2 )
If A is a real and symmetric matrix, then ๐จ = ๐ and
๐จ๐ = ๐จ, now compute[1]โฆ
[1]Peter V. OโNeil, โEigenvalues,Diagnolization and Special Matricesโ in Advanced Engineering Mathematics
5th edition. Birmingham, AL: B. Stenquist, 2003, ch. 8.3, pp. 354-362.
๐ธ ๐ ๐จ๐ธ = ๐ธ ๐ ๐จ๐ธ = ๐ธ ๐ ๐จ๐ธ is a 1x1 matrix (a number), and so
is the same as its transpose
โ ๐ธ ๐ ๐จ๐ธ = (๐ธ ๐ ๐จ๐ธ)๐ = ๐ธ ๐ ๐จ(๐ธ ๐ )๐ = ๐ธ ๐ ๐จ๐ธ
โ ๐ธ ๐ ๐จ๐ธ = ๐ธ ๐ ๐จ๐ธ = ๐( ๐1 2 + ๐2 2 + โฏ + ๐๐ 2 )
Therefore, the number ๐ธ ๐ ๐จ๐ธ , being equal to its conjugate,
is a real number. And ( ๐1 2 + ๐2 2 + โฏ + ๐๐ 2 ) is
certainly real. Therefore ๐ is real[1].
[1]Peter V. OโNeil, โEigenvalues,Diagnolization and Special Matricesโ in Advanced Engineering Mathematics
5th edition. Birmingham, AL: B. Stenquist, 2003, ch. 8.3, pp. 354-362.
๏ Let A be a real symmetric matrix. Then eigenvectors
associated with distinct eigenvalues are orthogonal.
๏ Let A be a real symmetric matrix. Then there is a real,
orthogonal matrix that diagonalizes A[1].
๏ Let A be a real, n x n symmetric matrix. Then its
eigenvector provide an orthogonal basis for n-space.
Therefore, if an eigenvalue is repeated by k times.
Then the eigenspace is of dimension k, and we can
find another set of orthogonal vector by linear
combination[2].
[1] Peter V. OโNeil, โEigenvalues,Diagnolization and Special Matricesโ in Advanced Engineering
Mathematics 5th edition. Birmingham, AL: B. Stenquist, 2003, ch. 8.3, pp. 354-362.
[2] M.D. Greenberg, "The Eigenvalue Problem," in Advanced Engineering Mathematics, 2nd ed. Upper
Saddle River, New Jersey: Prentice Hall, 1998, ch. 11.3, pp. 554-569.
Symmetric Matrix Examples
๏ฌ1 ๏ฝ 2
๏ฆ 3 0 ๏ญ2 ๏ถ
let A ๏ฝ ๏ง๏ง 0 2 0 ๏ท๏ท and get ๏ฌ2 ๏ฝ ๏ญ1
๏ง ๏ญ2 0 0 ๏ท
๏ฌ3 ๏ฝ 4
๏จ
๏ธ
๏ We can see that a real, symmetric matrix provides a set
of real eigenvalues. And the corresponding eigenvectors
are
๏ฆ 0๏ถ ๏ฆ 1๏ถ ๏ฆ 2 ๏ถ
๏ง1๏ท,๏ง 0๏ท,๏ง 0 ๏ท
๏ง ๏ท ๏ง ๏ท ๏ง ๏ท
๏ง 0 ๏ท ๏ง 2 ๏ท ๏ง ๏ญ1๏ท
๏จ ๏ธ ๏จ ๏ธ ๏จ ๏ธ
๏ These form an orthogonal set of vectors[1].
[1]Peter V. OโNeil, โEigenvalues,Diagnolization and Special Matricesโ in Advanced Engineering Mathematics
5th edition. Birmingham, AL: B. Stenquist, 2003, ch. 8.3, pp. 354-362.
๏ The orthonormal form is divided by its length and they
can be used as columns of an orthogonal matrix.
๏ฆ
1
2 ๏ถ
๏ง0 5 5 ๏ท
Q ๏ฝ ๏ง1 0 0 ๏ท
๏ง
๏ท
๏ง 0 25 ๏ญ 15 ๏ท
๏ง
๏ท
๏จ
๏ธ
๏ We can find Q-1 = QT, and A can be diagonalized by Q[1].
๏ฆ 2 0 0๏ถ
Q ๏ญ1 AQ ๏ฝ ๏ง๏ง 0 ๏ญ1 0 ๏ท๏ท
๏ง 0 0 4๏ท
๏จ
๏ธ
[1]Peter V. OโNeil, โEigenvalues,Diagnolization and Special Matricesโ in Advanced Engineering Mathematics
5th edition. Birmingham, AL: B. Stenquist, 2003, ch. 8.3, pp. 354-362.
Useful properties[1]
๏ If A, B are symmetric n × n matrices, then A+B is
symmetric.
๐ด + ๐ต ๐ = ๐ด๐ + ๐ต๐ = ๐ด + ๐ต
๏ If A, B are symmetric n × n matrices, then AB is not
symmetric.
๐ด๐ต ๐ = ๐ต๐ ๐ด๐ = ๐ต๐ด โ ๐ด๐ต
๏ If C is any n × n matrix. Then ๐ต = ๐ถ ๐ ๐ถ is a symmetric
matrix.
(๐ถ ๐ ๐ถ)๐ = ๐ถ ๐ (๐ถ ๐ )๐ = ๐ถ ๐ ๐ถ
๏ If D is a diagonal matrix, then D is symmetric.
[1]http://www.math.panam.edu/
BRAINBITE
Find which of followings is not the eigenvalue of the 4x4
matrix and its corresponding orthogonal eigenvector.
a. ๐ = 0, ๐ฃ = [1 0 0 0]๐
0 0
0 0
0 1 โ2 0
b. ๐ = 0, ๐ฃ = [0 0 0 1]๐
0 โ2 1 0
0 0
0 0
c. ๐ = โ1, ๐ฃ = [0 1 1 0]๐
d. ๐ = 1, ๐ฃ = [0 โ 1 โ 1 0]๐
e. ๐ = 3, ๐ฃ = [0 โ 1 1 0]๐
[1]Peter V. OโNeil, โEigenvalues,Diagnolization and Special Matricesโ in Advanced Engineering Mathematics
5th edition. Birmingham, AL: B. Stenquist, 2003, ch. 8.3, pp. 354-362.
answer : d
11.4 DIAGONALIZATION
Background
๏ Diagonal matrices have good properties for simplifying
calculations
๏ Exploit these properties by diagonalizing the matrix
A in ๐จ๐ = ฮป๐ or ๐จ๐ = ๐โฒ for DE
๏ Essentially, a change of base is required (๐ฑ = ๐๐) so
thatโฆ
๏ ๐จ๐ธ๐ = ๐ธ๐โฒ
๏ ๐โฒ = ๐ธโ1 ๐จ๐ธ ๐ = ๐ซ๐
Properties and Restrictions
๏ A is diagonalizable if and only if it has n LI
eigenvectors
๏ If the above condition is met, ๐ธ = {๐1 , ๐2 , โฆ , ๐๐ } (i.e.
the eigenvectors of A form a matrix Q)
๏ Symmetric matrices are always diagonalizable.
๏ ๐ธ๐ = ๐ธโ1 due to property of LI eigenvectors
Example
๏
๏
๏
๏
๐ฅ(๐ก)
3 5
๐ + ๐จ๐ = ๐ where ๐ =
and ๐จ =
๐ฆ(๐ก)
1 2
Make substitution of ๐ = ๐ธ๐
๐โฒโฒ + ๐ธโ1 ๐จ๐ธ๐ = ๐ where ๐ธโ1 ๐จ๐ธ๐ = ๐ซ๐
Solve for eigenvalues and eigenvectors to find D and Q
โฒโฒ
Example
๏ Eigenvalues are of A are ฮป1 = 4.79129 and ฮป2 =0.208712
2.79129
โ1.79129
๏ Eigenvectors are
and
1
1
4.79129
0
๏ Therefore, ๐ซ =
when ๐ =
0
0.208712
2.79129 โ1.79129
1
1
Example
๏ Using D, two uncoupled differential equations arise
(instead of coupled like before)
๐(๐)โฒโฒ + ๐๐ ๐ + ๐๐ฆ(๐ก) = ๐
๐(๐)โฒโฒ + ๐๐ ๐ + ๐๐(๐) = ๐ Diagonalization
๐โฒโฒ + ๐. ๐๐๐๐๐๐ = ๐
๐โฒโฒ + ๐. ๐๐๐๐๐๐๐ = ๐
๏ These equations are simple ODE and are solved using
the solution ๐ด sin(๐๐ก + ๐).
๏ ฯ can be solved for easily whereas A and ฯ are
constants of integration
Example
๏ Since we assumed ๐ = ๐ธ๐, now we can solve for the
real x(t) and y(t)
๐ฅ(๐ก)
๐ด1 sin(๐1 ๐ก + ๐1 )
๏
=๐ธ
๐ฆ(๐ก)
๐ด2 sin(๐2 ๐ก + ๐2 )
๐ฅ(๐ก)
2.79129
๏
=
๐ฆ(๐ก)
1
โ1.79129 ๐ด1 sin(๐1 ๐ก + ๐1 )
๐ด2 sin(๐2 ๐ก + ๐2 )
1
๏ x(t) and y(t) have been solved completely and easily
compared to not using properties of diagonal matrices
11.5 APPLICATIONS TO FIRST
ORDER SYSTEMS WITH
CONSTANT COEFFICIENTS
11.5 APPLICATIONS TO FIRST ORDER
SYSTEMS WITH CONSTANT COEFFICIENTS
๏ Consider an initial value problem
๏ In matrix form
๏ The solution to the differential equation is given by
๏ Where
A= coefficients of variables
Q= modal matrix =[e1,e2โฆ.en]
D= Diagonal matrix where jth diagonal elements
are jth eigenvalue of A.
๏ The solution can also be expressed of the form
Where
M.D. Greenberg, "The Eigenvalue Problem," in Advanced Engineering
Mathematics, 2nd ed. Upper Saddle River, New Jersey: Prentice Hall, 1998, ch. 11.3
Example
๏ Consider the equations
๏ Solution :
๏ Replacing the values of A,D,Q and Q-1 in the following
equation
๏ we get
11.6 QUADRATIC FORMS
๏ A (complex) quadratic form is an expansion
๐
๐=1
๐
๐=1 ๐๐๐ ๐ง๐ ๐ง๐ ,
in which each ๐๐๐ and ๐ง๐ is a
complex number[1]. For n=2, this is
๐11 ๐ง1 ๐ง1 + ๐12 ๐ง1 ๐ง2 + ๐21 ๐ง2 ๐ง1 + ๐22 ๐ง2 ๐ง2
๏ The quadratic form is real if each ๐๐๐ and ๐ง๐ is real, and
we usually write ๐ง๐ as ๐ฅ๐ , and the form is
๐
๐
๐=1 ๐=1 ๐๐๐ ๐ฅ๐ ๐ฅ๐ .
[1]Peter V. OโNeil, โEigenvalues,Diagnolization and Special Matricesโ in Advanced Engineering Mathematics
5th edition. Birmingham, AL: B. Stenquist, 2003, ch. 8.4, pp. 363-367.
๏ It is often convenient to write a quadratic form in a
matrix form. If A = [๐๐๐ ], ๐ง =
๐11
then ๐ง ๐ ๐ด๐ = (๐ง1 , ๐ง2 โฏ ๐ง๐ ) โฎ
๐๐1
=
๐
๐=1
๐ง1
๐ง2
โฎ
๐ง๐
โฏ ๐1๐
โฑ
โฎ
โฏ ๐๐๐
๐ง1
๐ง2
โฎ
๐ง๐
๐
๐=1 ๐๐๐ ๐ง๐ ๐ง๐ [1]
[1]Peter V. OโNeil, โEigenvalues,Diagnolization and Special Matricesโ in Advanced Engineering Mathematics
5th edition. Birmingham, AL: B. Stenquist, 2003, ch. 8.4, pp. 363-367.
Quadratic forms example
๏ Let ๐ด =
๐ฅ1 , ๐ฅ2
1 4
, then
3 2
1 4
3 2
๐ฅ1
๐ฅ2
= ๐ฅ1 2 + 7๐ฅ1 ๐ฅ2 + 2๐ฅ2 2
But we can also rewrite the quadratic form as ๐ฅ1 2 +
7
1
7
7
๐ฅ1
2
๐ฅ1 ๐ฅ2 + ๐ฅ2 ๐ฅ1 + 2๐ฅ2 2 = ๐ฅ1 , ๐ฅ2 7
2
2
2 ๐ฅ2
2
The advantage of latter form is that A is a symmetric
matrix[1].
[1]Peter V. OโNeil, โEigenvalues,Diagnolization and Special Matricesโ in Advanced Engineering Mathematics
5th edition. Birmingham, AL: B. Stenquist, 2003, ch. 8.4, pp. 363-367.
Classi๏ฌcation of The Quadratic Form[1]
Q = ๐ฅ ๐ ๐ด๐ฅ: A quadratic form is said to be:
๏ Negative de๏ฌnite: Q < 0 when x โ 0
๏ Negative semide๏ฌnite: Q โค 0 for all x and Q = 0 for
some x โ 0
๏ Positive de๏ฌnite: Q > 0 when x โ 0
๏ Positive semide๏ฌnite: Q โฅ 0 for all x and Q = 0 for
some x โ 0
๏ Inde๏ฌnite: Q > 0 for some x and Q < 0 for some other
x
[1]http://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=5&ved=0CGQQFjAE&url=http%3A
%2F%2Fwww.econ.iastate.edu%2Fclasses%2Fecon501%2FHallam%2Fdocuments%2FQuad_Forms_000.pdf
&ei=wp26TvrgMsiJsAKNzqXOCA&usg=AFQjCNGQ_OibQn6rhf0wrBTNSVMVOltoaQ&sig2=Y061Hf2_fqb
XqjUYyADczQ
Classi๏ฌcation example[1]
1 0 0
๏ For ๐ด = 0 2 0 , then
0 0 4
1 0 0 ๐ฅ1
๐ = ๐ฅ ๐ ๐ด๐ฅ = (๐ฅ1 , ๐ฅ2 , ๐ฅ3 ) 0 2 0 ๐ฅ2
0 0 4 ๐ฅ3
= ๐ฅ12 + 2๐ฅ22 + 4๐ฅ32
For any real vector ๐ฅ โ 0, that ๐ will be positive (so
called positive definite).
[1]http://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=5&ved=0CGQQFjAE&url=http%3A
%2F%2Fwww.econ.iastate.edu%2Fclasses%2Fecon501%2FHallam%2Fdocuments%2FQuad_Forms_000.pdf
&ei=wp26TvrgMsiJsAKNzqXOCA&usg=AFQjCNGQ_OibQn6rhf0wrBTNSVMVOltoaQ&sig2=Y061Hf2_fqb
XqjUYyADczQ
Graphical Analysis[1]
๏ Consider the indefinite matrix A is given by
โ2 2
2 2
๏ The quadratic form then is given by
โ2 2 ๐ฅ1
๐
๐ = ๐ฅ ๐ด๐ฅ = (๐ฅ1 , ๐ฅ2 )
2 2 ๐ฅ2
= โ2๐ฅ12 + 4๐ฅ1 ๐ฅ2 + 2๐ฅ22
๐ด=
= 4๐ฅ22 โ ( 2๐ฅ1 โ 2๐ฅ2 )2
๏ Then Q > 0 for some x and Q < 0 for some other x, so
called indefinite form.
[1]http://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=5&ved=0CGQQFjAE&url=http%3A%2F%
2Fwww.econ.iastate.edu%2Fclasses%2Fecon501%2FHallam%2Fdocuments%2FQuad_Forms_000.pdf&ei=wp26Tv
rgMsiJsAKNzqXOCA&usg=AFQjCNGQ_OibQn6rhf0wrBTNSVMVOltoaQ&sig2=Y061Hf2_fqbXqjUYyADczQ
Graphical Analysis[1]
๏ The graphic in 3-demension governed by Q = โ2๐ฅ12 +
4๐ฅ1 ๐ฅ2 + 2๐ฅ22 as follows
๏ Where it is clear that Q takes both positive and
negative values.
[1]http://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=5&ved=0CGQQFjAE&url=http%3A%2F%
2Fwww.econ.iastate.edu%2Fclasses%2Fecon501%2FHallam%2Fdocuments%2FQuad_Forms_000.pdf&ei=wp26Tv
rgMsiJsAKNzqXOCA&usg=AFQjCNGQ_OibQn6rhf0wrBTNSVMVOltoaQ&sig2=Y061Hf2_fqbXqjUYyADczQ
๏ In some problems involving quadratic forms,
calculations are simplified if we transform from
๐ฅ1 , ๐ฅ2 โฏ ๐ฅ๐ coordinate system to a ๐ฆ1 , ๐ฆ2 โฏ ๐ฆ๐ system
in which there are no mixed product terms. That is,
we want to choose so that ๐๐=1 ๐๐=1 ๐๐๐ ๐ฅ๐ ๐ฅ๐ =
๐
2
๐
๐ฆ
๐
๐
๐=1
๏ The ๐ฆ1 , ๐ฆ2 โฏ ๐ฆ๐ coordinates are called principal axes
for the quadratic form, where the rotation of axes is
used to eliminate mixed product terms in the
equation of conic[1].
[1]Peter V. OโNeil, โEigenvalues,Diagnolization and Special Matricesโ in Advanced Engineering Mathematics
5th edition. Birmingham, AL: B. Stenquist, 2003, ch. 8.4, pp. 363-367.
๏ Let A be a real symmetric matrix with eigenvalues ๐1 โฏ ๐๐ ,
and Q is an orthogonal matrix formed by their
corresponding eigenvectors that diagonalizes A. Then the
change of variables X=QY transforms ๐๐=1 ๐๐=1 ๐๐๐ ๐ฅ๐ ๐ฅ๐
to ๐๐=1 ๐๐ ๐ฆ๐ 2 [1].
๐
๐=1
๏ Proof:
๐
๐=1 ๐๐๐ ๐ฅ๐
๐ฅ๐ = ๐ ๐ ๐ด๐ =
๐๐)๐ ๐ด ๐๐ = (๐ ๐ ๐๐ ๐ด๐๐ = ๐ ๐ ๐๐ ๐ด๐ ๐
= ๐ฆ1 โฏ ๐ฆ๐
๐1
0
0
0
โฑ
0
0
0
๐๐
๐ฆ1
2 + โฏ+ ๐ ๐ฆ 2
=
๐
๐ฆ
1 1
๐ ๐
โฎ
๐ฆ๐
[1]Peter V. OโNeil, โEigenvalues,Diagnolization and Special Matricesโ in Advanced Engineering Mathematics
5th edition. Birmingham, AL: B. Stenquist, 2003, ch. 8.4, pp. 363-367.
Principal Axis Example
๏ Analyze the conic 4๐ฅ1 2 โ 3๐ฅ1 ๐ฅ2 + 2๐ฅ2 2 = 8
First write the quadratic form as ๐ฟ๐ ๐จ๐ฟ = 8, where ๐ด =
โ3
4
2
, then the eigenvalues of A are (6 ± โ13)/2.
โ3
2
2
By the principal axis theorem, there is an orthogonal
matrix Q that transforms the equation of the conic to
standard form (canonical form):
6+โ13
๐ฆ1 2
2
+
6โโ13
๐ฆ2 2
2
=8
[1]Peter V. OโNeil, โEigenvalues,Diagnolization and Special Matricesโ in Advanced Engineering Mathematics
5th edition. Birmingham, AL: B. Stenquist, 2003, ch. 8.4, pp. 363-367.
๏ This is an ellipse in the ๐ฆ1 , ๐ฆ2 plane. The figure[1]
shows a graph of this ellipse.
๐ = ๐๐ โ
๐ = ๐ โ1 ๐ = ๐๐ ๐
[1]Peter V. OโNeil, โEigenvalues,Diagnolization and Special Matricesโ in Advanced Engineering Mathematics
5th edition. Birmingham, AL: B. Stenquist, 2003, ch. 8.4, pp. 363-367.