Linear Algebra and Advanced Calculus: Somitra Sanadhya

Linear Algebra and Advanced Calculus
Somitra Sanadhya
IIIT-Delhi
somitra@iiitd.ac.in
Lecture 20, Monsoon semester 2010

Oct 25, 2010
Lecture 20
Eigenvalues of a Symmetric matrix
Eigenvalues of a Symmetric matrix

Let λ1 and λ2 be two distinct eigenvalues of a symmetric
matrix A.
Then the corresponding eigenvectors are orthogonal.
Proof: Ax1 = λ1 x1 and Ax2 = λ2 x2 . Multiply both sides of
the first one by xT2 , to get xT2 Ax1 = λ1 xT2 x1 .
Look at the LHS of the newly formed equality.
xT2 Ax1 = xT2 AT x1 = (Ax2 )T x1 = λ2 xT2 x1 .
Hence λ1 xT2 x1 = λ2 xT2 x1 . Since λ1 6= λ2 , we have xT2 x1 = 0.
The eigenvectors of a symmetric matrix can be chosen to be
orthogonal, even when the eigenvalues are not distinct.
Example: I2 .
Lecture 20
Spectrum of a matrix
Spectrum of a matrix
Eigenvectors are the “Spectrum” of the matrix.
Give a lot of information about the matrix.
For a symmetric matric, one can write A = QΛQ−1 = QΛQT
where Q is an orthogonal matrix.
The above step is A = λ1 q1 q1T + λ1 q2 q2T + . . . + λ1 qn qnT
A symmetric matrix A can be expressed as the sum of rank 1
matrices, each of which is constructed from mutually
orthogonal columns. (These directions are called principal
axes.)
The last statement is one way of expressing the “spectral
theorem”.
Lecture 20
Eigenvectors as the basis
Eigenvectors as the basis

Let the independent eignevectors of an n × n matrix A be
x1 , x2 , . . . , xn .
Any vector x ∈ Rn can be uniquely expressed as a linear
combination of the eigenvectors
x = c1 x1 + c2 x2 + . . . + cn xn .
(Full set of independent) Eigenvectors are a basis of Rn .

Lecture 20
Solving recurrrence equations
Solving recurrrence equations

Let uk+1 = Auk and u0 is given. Aim: Solve for uk .
Substitution gives uk = Ak u0 . But finding Ak maybe difficult.
Write u0 as a linear combination of eigenvectors of A. That
is, u0 = Sc = c1 x1 + c2 x2 + . . . (where S is the eigenvector
matrix and c is a vector of coefficients).
Now see that Ak u0 = c1 λk1 x1 + c2 λk2 x2 + . . ..
Each eigenvector goes in its direction independently.
(Stretches or compresses).
Lecture 20
Fibonacci Series
Fibonacci Series
0, 1, 1, 2, 3, 5, 8, 13, 21, . . .
Fk+2 = Fk+1 + Fk and F0 = 0, F1 = 1.
Can be written as
1 1 Fk+1
uk+1 = Auk where A = and uk =
1 0 Fk
Eigenvalues and eigenvectors
of A:
√ √
1+ 5 λ1 1− 5 λ2
λ1 = 2 , x1 = and λ2 = 2 , x1 =
1 1
u0 = Sc gives [1 0]T = [x1 x2 ][c1 c2 ]T . We can solve for
c1 , c2 .
uk = c1 λk1 x1 + c2 λk2 x2 .
Lecture 20
Stability, ...
Stability, ...
For a matrix A, having eigenvalues λi , i = 1, 2, . . .
If A is diagonalizable, then A = SΛS −1 and hence
Ak = SΛk S −1 .
Consider the behaviour of Ak as k increases.
Three cases
1 Blow-up (some elements of Ak → ∞)
2 Steady state (Ak+1 does not differ much from Ak as k
increases)
3 Stable (irrespective of the initial value, Ak always converges to
the same matrix)
Note that this discussion holds only for diagonalizable
matrices.
Lecture 20
Differential equation
du1
dt = −u1 + 2u2 ,
du2
dt = 2u1 − u2 and the initial condition
u1 (t = 0) = 1, u2 (t = 0) = 0.

du −1 2
In other words, = Au where A =
dt and
2 −1

u1
u=
u2
Find eigenvalues
and eigenvectors for A:
2 1
λ1 = 0, x1 = and λ2 = −3, x2 =
1 −1
General solution is u(t) = c1 eλ1 t x1 + c2 eλ2 t x2 .
Find c1 , c2 from initial condition. c1 = c2 = 31 .
Lecture 20
General idea: du
dt = Au. Let S be the eigenvector matrix of A
and let u = Sv (substitution: change of variables).
The equation changes to S dv
dt = ASv. This gives
dv −1 ASv.
dt = S
dv
If A is diagonalizable, then S −1 AS = Λ. Hence, dt = Λv.
Much easier to solve since the equations have decoupled.
Solution is v(t) = eΛt v(0).
Reverting to the original variable: u(t) = SeΛt S −1 u(0).
Lecture 20
Exponent of a matrix
Exponent of a matrix
eAt = I + (At) + (At)2 /2! + (At)3 /3! + . . .
From the previous example: we can show that eAt = SeΛt S −1 .
Proof idea:
1 eAt = I + (At) + (At)2 /2! + . . .
2 Use A = SΛS −1
3 Get eAt = S (I + (Λt) + (Λt)2 /2! + . . .) S −1
Why do this ? eΛt is much easier to compute than eAt
 λt 
e 1 ... 0
eΛt =  ... ..
.
.. 

. 
0 0 e ntλ
Lecture 20
Solving higher order differential equation
Solving higher order differential equation

Consider the second order eqn: y ′′ + ay ′ + b = 0
′
′ y
It can be written as Y = AY where Y = and
y

−a −b
A=
1 0
Now use techniques similar to what we already did.
In general, an equation of order 5 can be converted to a 5 × 5
matrix A which has I4 as a submatrix.

Linear Algebra and Advanced Calculus: Somitra Sanadhya

Uploaded by

Copyright:

Available Formats

Linear Algebra and Advanced Calculus: Somitra Sanadhya

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Linear Algebra and Advanced Calculus: Somitra Sanadhya

Uploaded by

Copyright:

Available Formats

Linear Algebra and Advanced Calculus

Linear Algebra and Advanced Calculus

Lecture 20, Monsoon semester 2010

Eigenvalues of a Symmetric matrix

Eigenvectors as the basis

(Full set of independent) Eigenvectors are a basis of Rn .

Solving recurrrence equations

Solving higher order differential equation

You might also like