5) Contraction mappings and fixed point theorems Flashcards
fixed point
suppose that (X, d) is
a metric space and f : X → X a self-map. Then a point x ∈ X is called fixed point of f
if f(x) = x. For example the function cos defines a self-map on the interval [0, 1], and by
starting with x1 = 0 and inductively computing xn+1 = cos xn one converges to the value
roughly 0.739085 which is a fixed point of cos, i.e. solves the equation cos(x) = x
previously discussed Stone Weirstrass theorem
That a collection, space of continuous functions on a compact set will be separable-
proof uses that we can approx any continuous functions by a polynomial
And any polynomial with real coefficients can be nicely approximated by poly with rational coefss as rationals approximate reals
There are a countable number of polynomials with rational coefss,
so we have a countable everywhere dense set which was required for separability of this space
Definition 5.1 (Contraction Mapping).
Let (X, d) be a metric space. Then a map f : X → X is called contraction if there exists a constant C < 1 such that d(f(x), f(y)) ≤ Cd(x, y).
we have a strict inequality here
(clearly any contraction map will be uniformly continuous- if not contunous then cant be a contraction!(disprove))
(clearly any contraction map will be uniformly continuous)
For all x and for all ε there exists a δ_ε>0 s.t. d(x,y) <δ_ε implies d(f(x),f(y))<ε
A contraction is a special kind of uniformly continuous maps, here we take δ = Cε or ε
contraction mapping idea
We look for a fixed point x s.t f(x)=x
E.g for cos x sketch y=x and take say x_1=1, cos(1) then use x_2=cos(2) to find points (spiral formed until fixed point is reached).
x~= lim n tends to infinity of x_n = lim n tends to infinity of x_n+1 = lim as n tends to infinity of cos(x_n) = cos (x~)
sometimes we might have the function dropping much faster, the similar procedure wouldnt work, spirals out and fixed point doesnt exist, there is no convergence. Here what is different is that the derivative |f’| > 1 here
Slope tells us we will have con traction of argument or not
contraction mapping
y=ax+b on a:
consider y=x and line y=ax+b. If |a|=1 we will have a closed square
|a|>1 divergence no fixed point
|a|<1 fixed point
y_1=ax_1+b
y_2=ax_2+b
|y_2-y_1| = |a||x_1-x_2|
Theorem 5.2 (Banach Fixed Point Theorem)
Suppose that f : X → X is a contraction on a complete metric space (X, d). Then f has a unique fixed point y.
Moreover, for any x ∈ X the sequence (x_n) defined recursively by x_{n+1} = f(x_n), x_1 = x,
converges to y.
(we shrink towards the point using this recursive relation)
e.g. unit circle rotated by pi/2
consider unit circle
if we rotate circle by an angle no point will remain in place
But the distance between the mapped points is preserved
(we will say not bigger than)
d(f(x),f(y)) ≤ d(x,y)
Theorem 5.2 (Banach Fixed Point Theorem) PROOF
steps: **
Existence of fixed point
Uniqueness of fixed point
this fixed point can be obtained by result of approximation by iterative procedure
**
1)
Assume we have two fixed points x and y. But d(f(X),f(y) = d(x,y) ≤ Cd(x,y)
but since C ≤ 1 we must have d(X,Y) = 0 otherwise inequality not possible, so points coincide.
2)
if f is a contraction we have uniform continuity, in particular its continuous. If we have a convergent sequence x_{n+1} = f(x_n) is convergent to x~
Then x = (limn→ ∞)(x_n)
= (limn→ ∞)(x_{n+1})
by continuity
= = (limn→ ∞)(f(x_n))
=f( = (limn→ ∞)(x_n))
=f(x~)
Why x_n is cauchy?
using series 1+q+q^2+…. = (1-q^(n+1)/(1-q)
the limit of partial sums up to n will exist if |q|<1
limit as n tends to infinity of this will be 1/(1-q) if |q|<1
wlog m>n then m=n+k
d(x_n,x_m) ≤d(x_n,x_(n+1)) + ….d(x_m-1,x_m)
by induction
As d(x_{n+1},x_{n+2}
= d(f(x_n),f(x_{n+1}) ≤Cd(x_n, x_{n+1)}
….
d(x_{n+k}, x_{n+k+1}) ≤ C^k d(x_n, x_{n+1})
By triangle inequality
d(x_{n+k},x_{n+k+1})
≤ d(x_n, x_{n+1)(1 +C + C^2 +…C^k)
≤ C^{n-1}d(x_1, x_2)(1 +C + C^2 +…C^k)
≤ (C(n-1) / (1-c) d(x_1,x_2)< ε
Ensure this happens by:
Taking ε such that
C^{n-1} <
[ ε(1-C)]/[d(x_1,x_2)]
(inequality swaps as logC<0 since C< 1)
n-1 > log([ ε(1-C)]/[d(x_1,x_2)]) /log(C)
N> [log([ ε(1-C)]/[d(x_1,x_2)]) /log(C) ] + 1
Corollary 5.3. composition of map several times that wasn’t originally a contraction
Suppose that (X, d) is a complete metric space and f : X → X a map
such that f^ n is a contraction for some n ∈ N. Then f has a unique fixed point.
proof: Since f^n is a contraction it has a unique fixed point x ∈ X, i.e.
f ◦ f . . . ◦ f(x) (n−times) = x.
Now note that
f^n(f(x)) = f^n ◦ f(x) = f^{n+1}(x) = f ◦ f^n(x) = f(f^n(x)) = f(x)
and therefore f(x) is also a fixed point of f^n . By uniqueness we must have f(x) = x
Jacobian
We will sue this to verify if a map is a contraction for several variables
Partial derivatives matrix
f: R^n to R^m
( ∂f_1/ ∂x_1…….. ∂f_1/ ∂x_m)
(………………………)
( ∂f_m/ ∂x_1…….. ∂f_m/ ∂x_m)
square matrix and compute the determinant
∥∂f(x)∥≤∥∂f∥*∥x∥
j
(here norm as a linear operator)
Theorem 5.4 (Mean Value Inequality).
Used for contractions
Suppose that U ⊂ R^m is an open set with
convex closure overline(U) and let f : overline(U) → R^m be a C^1
-function. Let df be the total derivative
(or Jacobian) understood as a function on U with values in m × m-matrices. Suppose that ∥df(x)∥ ≤ M for all x ∈ U. Then f : U → R
m satisfies
∥f(x) − f(y)∥ ≤ M∥x − y∥
for all x, y ∈ overline(U)
The Mean Value Theorem for Integrals
he Mean Value Theorem for Integrals states that a continuous function on a closed interval takes on its average value at the same point in that interval. The theorem guarantees that if f(x) is continuous, a point
c exists in [a,b] such that the value of the function at c is equal to the average value off(x) over [a,b]
ie there exists a c in [a,b] s.t
f(c) = (1/(b-a)) . ∫_[a,b] f(x).dx
*since continuous on[a,b]closed interval is bounded
m<f(x)<M
m(b-a)<∫_[a,b] f(x).dx< M(b-a)
*by IVT there is a c s.t f(c) =z for f(z)< z< f(b)
norm as linear operator
∥∂ f(x)∥ ≤ ∥∂ f∥ ∥x∥
convex domain assumed
convex
for two points in our interval joined by straight line this interval is in our domain
γ(t) = tx + (1 − t)y
d(f(x),f(y)) can be estimated along this interval which reduce our number of variables to 1
PROOF OF MVI
Using convex domain, for x,y in U we have γ(t) = tx + (1 − t)y
γ((1)=x and γ(0)=y
(d/dt)γ(t) = x − y.
f(x)-f(y) = ∫[0,1] d/dt(f(γ(t)).dt
= ∫0,1 . (dγ/dt)(t).dt
ftoc
by MVT
using the triangle inequality (useful for Riemann integrals as they are limits of finite sums)
df is the jacobian
so f(x)-f(y) is the integral of product of jacobian and (x-y) (x-y is the direction of the line prev)
∥f(x) − f(y)∥ ≤
∫[0,1] ∥(df).(dγ/dt)(t) ∥.dt
≤
M*
∫[0,1] ∥x-y ∥.dt = M∥x-y ∥
triangle inequality on a sum using norms
∥sum of x_k ∥
applied to integrals
∥sum of x_k ∥
≤ sum of ∥x_k∥
∥ ∫f(t).dt∥ ≤ ∫ ∥ f(t).dt∥
example 5.5 Consider the map
f:R^2 ⊃ overline(B_1(0)) →overline(B_1(0))
(x,y)→(x²/4 + (y/3) + (1/3) , y²/4 - (x/2))
jacobian df
find the norm of df
df
(x/2 1/3)
(-1/2 y/2)
Finding the norm of linear operator is challenging
∥T∥ =
sup_(∥x∥=1) [ ∥T(x)∥]
but we use HS norm:
df * df^T
and take trace = x²/4 + 1/9 + y²/4 + 1/4
take sqrt < 1
Therefore f is a contraction. We can find the fixed point by starting, for example, with the point (0, 0) and iterating.
Hilbert -schmidt norm
in an exercise we found that
∥T∥ ≤∥T∥_HS
∥A∥_HS =
(tr(A∗A^T))^0.5
initial value problem (IVP)
dy/dt = f(t, y),
y(t_0) = y0,
family of solutions, initial gives condition for a single or multiple solutions?
For f:K → R
K = [T₁, T₂] × [L₁, L₂] in R^2
y: [T₁,T₂]→ R, t → y(t)
and y_0 in [L₁,L₂]
t_0 in [T₁,T₂]
diagrams
family of functions and vector fields and integral curves for next examples
boundaries from the initial conditions
for e.g. 5.8 intitial values give different behaviours
5.9 solution not unique for some different initial values, all cross xaxis
Hence, there are two fundamental questions here: existence and uniqueness of solutions. The
following theorem is one of the basic results in the theorem of ordinary differential equation
and establishes existence and uniqueness under rather general assumptions
Example 5.6. Let
f(t, x) = x and
y₀ = 1,
t₀ = 0.
Then the initial value problem is
dy/dt = y,
y(0) = 1
we know there will be a unique sol y(t)=exp(t)
Example 5.7. Let
f(t, x) = x² and
y₀ = 1,
t₀ = 0.
Then the initial value problem is
dy/dt = y²,
y(0) = 1
we know there will be a unique sol y(t)= 1/(1-t)
Example 5.8. Let
f(t, x) = x²-t and
y₀ = 1,
t₀ = 0.
Then the initial value problem is
dy/dt = y²-t
y(0) = 1
One can show that there exists a solution for small |t|, however this solution cannot be expressed in terms of elementary functions