Functional Analysis by R.
Vittal Rao
Lecture 9: Contraction Mapping - June 20, 2012
We denote this by EF (The elastic limit of mapping F).
ρ(F(x ),F(x ))
∴ We have ρ(x11 ,x2 ) 2 ≤ EF , ∀x1 , x2 ∈ X.
9.1 Role of Completeness Suppose EF < 1, then the mapping F is called a contraction map
i.e.,
Let (X, ρ) be a metric space. We look at a function
ρ(F(x1 ), F(x2 )) ≤ EF ρ(x1 , x2 ), 0 ≤ EF < 1
F : X −→ X
Definition 9.1 Contraction Map
A function F : (X, ρ) −→ (X, ρ) is called a contraction map
& on (X, ρ) if ∃ a real constant EF such that 0 ≤ EF < 1,
ρ(F(x1 ), F(x2 )) ≤ EF ρ(x1 , x2 ).
!"#$ !"&'#$( An example:
!"#% !"&'#%( Let (X, ρ) = l2N (C)
If
x = (x1 , x2 , ..., xN ) ∈ l2N (C)
We define the ratio y = (y1 , y2 , ..., yN ) ∈ l2N (C)
ρ(F(x1 ),F(x2 )) then 1/2
ρ(x1 ,x2 ) N
2
X
ρ2 (x, y) = x j − y j
Where ρ(x1 , x2 ) is the distance between x1 and x2 .
j=1
ρ(F(x1 ), F(x2 )) is the distance between F(x1 ) and F(x2 ).
Suppose ∃ K ≥ 0 such that Let M be a fixed N × N complex matrix.
Define
ρ(F(x1 ),F(x2 ))
ρ(x1 ,x2 ) ≤ K, for any x1 , x2 ∈ X.
F : l2N (C) −→ l2N (C)
We find as
ρ(F(x1 ),F(x2 ))
lubx1 ,x2 ∈X and x1 ,x2 ρ(x1 ,x2 ) ≤K F(x) = Mx ∀ x ∈ l2N (C).
1
9.1 Role of Completeness
Under some conditions this F will become a contraction map. Definition 9.2 Fixed Point
Let λ1 , λ2 , ..., λN be the eigenvalues of M (they may be complex Let F : (X, ρ) −→ (X, ρ) then a point x ∈ (X, ρ) is called a fixed
and may be repeated). point for F if
Let the spectral norm or spectral radius of M is defined as
F(x) = x.
n o
kMksp = max λ j
1≤ j≤N A sufficient condition for F to have a fixed point is that it be a
contraction map. This is called contraction mapping theorem.
then we can show that
ρ2 (F(x), F(y)) = ρ2 (Mx, My) ≤ kMksp ρ2 (x, y) Theorem 9.1 Contraction Mapping Theorem
Let (X, ρ) be a complete metric space and F : (X, ρ) −→ (X, ρ) a
and also note that contraction map then F HAS a UNIQUE fixed point.
ρ2 (Mx,My)
lubx,y∈l2 (C) and x,y ρ2 (x,y) = kMksp Proof:
N
Let x0 be any element in (X, ρ).
If the spectral radius of M, kMksp < 1 then the map Then we let x1 = F(x0 )
and in general having found
F : l2N (C) −→ l2N (C)
x1 , x2 , ..., xn
defined as
we find
F(x) = Mx ∀ x ∈ l2N (C)
xn+1 = F(xn ).
is a contraction mapping on l2N (C). Thus we get a sequence {xn } in (X, ρ) defined as
If y ∈ l2N (C) is any fixed vector then the map
xn+1 = F(xn )
F(x) = Mx + y
with x0 is an arbitrary point.
is also a contraction map, because Strategy:
Claim 1: If F is a contraction map then {xn } is a Cauchy sequence in
ρ2 (F(x1 ), F(x2 )) (X, ρ).
= ρ2 (Mx1 + y, Mx2 + y)
Claim 2: Since (X, ρ) is complete ∃ x ∈ (X, ρ) such that xn −→ x.
= ρ2 (Mx1 , Mx2 )
≤ kMksp ρ2 (x1 , x2 ) Claim 3: F(xn ) −→ F(x) in (X, ρ).
Therefore if M is a fixed N ×N complex matrix such that the kMksp < Claim 4: F(x) = x i.e., x is a fixed point.
1 then for any fixed y ∈ l2N (C), the function F(x) = Mx + y is a
contraction mapping on l2N (C). Claim 5: x is the only fixed point.
2
9.2 Application of Contraction Mapping Theorem to Matrix Problems
Now consider
Proving claims:
NOTE: We prove claims backwards i.e., we assume earlier claims ρ(xn+1 , xn ) = ρ(F(xn ), F(xn−1 ))
are true then we prove current claim. ≤ EF ρ(xn , xn−1 )
Claim 5: Suppose x1 , x2 are two fixed points. ≤ E2F ρ(xn−1 , xn−2 )
...
Exercise: Show ρ(x1 , x2 ) = 0.
...
Hint: Use x1 = F(x1 ), x2 = F(x2 ) and F is a contraction map. ...
≤ EnF ρ(x1 , x0 )
Claim 4: We have xn+1 = F(xn ), now let n −→ ∞
Hence (9.1) becomes
xn+1 −→ x
F(xn ) −→ F(x) ≤ (EnF + En+1
F
+ ... + Em−1
F
)ρ(x1 , x0 )
∴ F(x) = x.
(EnF + En+1 + ... + Em−1 ) can be made < for n, m ≥ N as it is equals to
Claim 3: To prove claim 3 show that ρ(F(xn ), F(x)) can be made F F
Sm − Sn , where Sk is the kth partial sum of the convergent geometric
small.
series
ρ(F(xn ), F(x)) ≤ EF ρ(xn , x)
1 + EF + E2F + ...
Since xn −→ x, given 1 > 0 we can get N1 such that n ≥ N1 ⇒
ρ(xn , x) < 1 .
For > 0, choose 1 = /EF .
∴ ρ(F(xn ), F(x)) < for n ≥ N1 .
∴ F(xn ) −→ F(x). 9.2 Application of Contraction Mapping
Theorem to Matrix Problems
Claim 2: xn −→ x because by claim 1, it is Cauchy sequence
and space is complete. Let A be a N × N complex matrix.
Suppose we look at the linear system
Claim 1: To show that {xn } is a Cauchy sequence we have to
show that xn and xm get closer and closer i.e., we have to estimate Ax = y (9.2)
what ρ(xn , xm ) is going to be for large n and m. If n = m then there
Where y is a given fixed N component vector.
is nothing to prove.
Find solution by iterative method.
So let us assume n , m and without loss of generality assume
The setup: We try to get the solution x as the fixed point of
m > n,
contraction map on l2N (C).
ρ(xn , xm ) ≤ ρ(xn , xn+1 ) + ρ(xn+1 , xm )
≤ ρ(xn , xn+1 ) + ρ(xn+1 , xn+2 ) + ρ(xn+2 , xm ) The standard strategies:
≤ ρ(xn , xn+1 ) + ρ(xn+1 , xn+2 ) + ... + ρ(xm−1 , xm ) (9.1) 1. Jacobi method.
3
9.2 Application of Contraction Mapping Theorem to Matrix Problems
2. Gauss-Seidel method. written as
3. Relaxation method. ωAx = ωy
(ωD + ωL + ωU)x = ωy
The Basic Idea:
Dx + ((ω − 1)D + ωL + ωU)x = ωy
A can be written as
(D + ωL)x = −((ω − 1)D + ωU)x + ωy
A=D+L+U x = −(D + ωL)−1 ((ω − 1)D + ωU)x
+(D + ωL)−1 ωy
then the system (9.2) becomes
x = F(x)
(D + L + U)x = y. Where F(x) = Rω x + Y.
Where Rω = −(D + ωL)−1 ((ω − 1)D + ωU) and Y = ω(D + ωL)−1 y.
1. Jacobi Method:
If kRω ksp < 1 then we get x as the fixed point for the contraction
map F.
Dx = −(L + U)x + y
The idea is to choose ω which gives lowest value for kRω ksp and
x = −D−1 (L + U)x + D−1 y apply the scheme.
x = F(x)
Where F(x) = Jx + Y
Where J = −D−1 (L + U) and Y = D−1 y.
Note that D−1 is guaranteed to exist.
If kJksp < 1 then we get x as the fixed point for the contraction map F.
2. Gauss-Seidel Method:
The system (9.2) can be written as
(D + L)x = −Ux + y
x = −(D + L)−1 Ux + (D + L)−1 y
x = F(x)
Where F(x) = Gx + Y
Where G = −(D + L)−1 U and Y = (D + L)−1 y.
If kGksp < 1 then we get x as the fixed point for the contraction map F.
3. Relaxation Method:
Let ω , 0 be any complex number, then the system (9.2) can be
