The Weierstrass Approximation

Introduction
 Why do we make assumptions of “bounded interval” and
“continuous function” ?

Definition ● Bounded: helps ensure uniform convergence

● Continuous: discontinuous functions have abrupt

jumps/breaks that can’t be approximated by
Definition: Let f(x) be a continuous real-valued function
polynomials.
defined on a closed interval [a,b] where a<b. Then, for any
positive number ε > 0, there exists a polynomial P(x) such that
for all x in the interval [a,b], the absolute difference between
f(x) and P(x) is less than ε.

∀ x ∈ [a,b] ,∃ P(x) (a polynomial) such that ∣ f(x)−P(x)∣

< ε.

In simple words: any continuous function on a bounded

interval can be uniformly approximated by polynomial
functions.

Figure 1: graph showing approximation of polynomials to a given function

Dirac Delta Function ● In the context of convolution, the Dirac delta
function plays a special role as an identity
Definition: a function on the real line which is zero everywhere element.
except at the origin, where it is infinite.

● f * δ(x) = f (x - t) δ(t) dt = f (x) since δ(t) = 0

The Dirac delta function is not a traditional function that can
be evaluated at a point like regular functions. Instead, it is a except at t = 0.
distribution, which means its primary use is in integrals.

Figure 2: graph representing the dirac delta function

Lebesgue’s Proof :

1. Dirac Delta Function

2. Landau Kernel
Can we construct sequences
that behave like the dirac
delta as the limit tends to
infinity?...
Dirac Sequences:
While you cannot construct a sequence of regular functions that exactly replicate the Dirac delta function, you
can create sequences of functions that approximate its behavior as a limit. These sequences are called "Dirac
Sequences."

The properties of these approximations are as

follows:

δε(x) is smooth and continuous.

δε(x) approaches zero as ε approaches zero.
The integral of δε(x) over the entire real line is 1 for
each ε.
As ε approaches zero, the sequence {δε(x)}
becomes increasingly concentrated around the
origin and approaches the behavior of the Dirac
delta function. In the limit as ε tends to zero, δε(x)
converges to δ(x) in the sense of distributions.
Dirac Sequences:
What we will now do is find a sequence of functions (Kn) which approximate the δ-function. The
sequence (Kn* f) will then approximate δ * f = f.

The n-th Landau kernel function Kn= cn(1 - x2)

n for x ∈ [-1, 1]
and 0 otherwise, where the coefficient (cn) is chosen so that the
integral of the function is 1.
Dirac Sequences:

Using Integration by parts, we find that;

Can we construct sequences
that behave like the dirac
delta as the limit tends to
infinity?...
Proof I:
Lemma 1: If f is a continuous function on the interval [-1, 1] then Kn * f is a
polynomial

Kn* f (x) = (-1 -> 1)∫Kn(x - t)f (t) dt

Kn is a polynomial and so Kn(x - t) can be expanded as:

Kn(x-t)= g0(t) +g1(t) ... + g2n(t)x2n

and so the integral is a polynomial in x.

Proof I:
Lemma 2: The sequence (Kn * f)→ f in d∞.

Methodology: Show that Kn has most of its area concentrated near x=0.

Step 1: find out how big cn is.

Proof I:
The focus then shifts to analyzing the behavior of Kn away from the origin. It's
shown that the integral of Kn(t) from -1 to 1 is bounded by (n+1)/2 multiplied by
(1 - δ^2)^n. The choice of δ here ensures that Kn(t) is small for t away from 0.

It is observed that If r = 1 - δ2 then (n + 1)rn--> 0 as n→ ∞. This implies that as n becomes

larger, the Landau kernel Kn becomes more concentrated around the origin (x = 0).
Proof I:
Take a continuous function f and let it be bounded by M.

Now on [-1, -δ] and on [δ, 1] we have Kn(t) is small if we choose δ small. In fact, we can
choose δ so that Kn(t) < ε/M here and then the first and third integrals are < ε.

For the middle integral, |f (x) - f (x - t)|Kn(t) dt ≤ εKn(t) dt < ε since Kn(t) dt = 1.

Thus |f (x) - Kn* f (x)| is small when n is large and we have our convergence. This completes
the

proof of the Weierstrass approximation theorem.

Bernstein’s Proof
Bernstein Basis Polynomials ● In more technical terms, each Bernstein polynomial
represents the contribution of one control point to the
overall shape. By adjusting these contributions, you
can create a wide variety of curves and shapes that
Bernstein basis polynomials, also known as Bernstein look really nice and smooth.
polynomials, are a family of polynomial basis functions defined
on the closed interval [0, 1].

Properties:
1) Non-negativity: Bn(x) is non-negative on the interval
[0, 1],
2) Partition of Unity: The sum of all Bn(x) from k = 0 to n
is equal to 1 for all x in [0, 1]. In other words, they
form a partition of unity, ensuring that the combination
of all basis functions equals 1.
3) Degree: The degree of the nth Bernstein polynomial is
n.
 Bézier Curves
ic curves defined in terms of ● In more technical terms, each Bernstein polynomial
stein represents the contribution of one control point to the
ynomials. Given points in the plane , the degree Bézier overall shape. By adjusting these contributions, you
termined by the Pk is: can create a wide variety of curves and shapes that
look really nice and smooth.

ned by a set of control points, which are points that influence the shape of the curve.

tein polynomials determine how much influence each control point

curve at different values of t. As t varies from 0 to 1, the curve
ransitions from one control point to another, creating a smooth shape.
 Bézier Curves Example 1:

Let f(x)=√x and n=3.

Multiplying out and collecting terms:

The linear term of dominates, but higher order terms pull it up a

bit towards the function. As n increases, Bn(f) approaches f in a
uniform way.
Proof II:
Divide and Conquer Approach: Bernstein's main idea is to divide the problem into two parts:

Part 1: Focus on a small interval around that specific point where the Bernstein polynomial is small. This
part relies only on the uniform continuity of the function we want to approximate.

Part 2: Look at the intervals to the left and right of Part 1, where the Bernstein polynomial is large. This
part is solved by including aspects of probability theory. s. If is the probability of an event is x=p, then
Bk,n(p) is the probability that the event occurs exactly times in Bernoulli trials. This was Bernstein's
take-off point and he cited Bernoulli's probabilistic results to crimp the Bk,n(p).
Proof II:
Binomial Coefficients: Bernstein employs the binomial coefficients to manipulate the Bernstein polynomials'
properties. The key idea is to use the binomial coefficients to control the behavior of the Bernstein
polynomials and make them behave in a way that approximates the target function effectively.

Probabilistic Analogy: Bernstein draws an analogy between the behavior of the Bernstein polynomials in
these subintervals and the outcomes of Bernoulli trials. He uses the binomial coefficients to guide this
analogy. Essentially, he treats the coefficients as probabilities, and the Bernstein polynomial becomes a kind
of weighted sum of "success" and "failure" terms.

Crimping: This is where the probabilistic reasoning comes into play. Just as in Bernoulli trials, where you
can control the number of successes, Bernstein manipulates the Bernstein polynomial in subintervals (2) to
"crimp" or control its behavior using the properties of binomial coefficients. This ensures that the
approximation error is controlled and small.
Proof II:
1. Absorption Identity:
Proof II:

(sum is re-indexed with

j=k−1)

.
 Proof II:
(3) is proved similarly using absorption twice.

Divide inequality at the last step by n^2:

Proof II:
Let x belong to [0,1] and dirac delta>0 and consider the sum of all bernstein polynomials with k such that
|k/n-x|>=delta.

This inequality is true only for k such that, |k/n-x|>=delta. So if, |f(x)|<M on [0,1] then,
Proof II:
For k with k/n closer to x the uniform continuity of f can be used as follows. Let Let ϵ be given and choose
δ such that f(u)−f(v)|<ϵ/2 when |u−v|<δ throughout [0,1]. So for k with |k/n−x|<δ

Combining (4) and (5) we get,

This concludes our proof.