Uniform Approximation by Polynomials

Content:

Conditions:

1, $f\in C[a,b].$ That is a very weak condition.

Conclusions:

1, ∃ { P n } ⇉ f \exist\{P_n\}\rightrightarrows f ∃{Pn​}⇉f, where $P_n$ are polynomials with rational coefficients.

---

Examples:

1, This says stage is important.

$\frac{1}{x}$ cannot be approximated by polynomials on $(0,1]$.

---

Then we give three proofs that go along different main ideas.

---

---

Broken-Line Revamp

By Lebesgue, who uses a very, very straightforward method to prove this theorem.

---

Essence:

1, Since $f$’s uniformly continuous, we know the broken-lines, constructed by connecting all end-points in partitions, will uniformly converge to $f$.

2, Broken-line is actually a morph of $|x|$.

3, $|x|$ can be uniformly approximated by polynomials.

Idea:

We just need to revamp a broken-line to a polynomial.

Tricks:

1, $u_0(x):\equiv 0,u_{n+1}(x):=u_n(x)+\frac{x^2-u_n^2(x)}{2}.u_n(x)\rightrightarrows |x|,n\to +\infty .$ Thus Essence 3 is utilised.

2, Remember $\text{ReLU}$? Now we define $\text{ReLU}$ at $x_0$:
$$\text{ReLU}(x_0):=\frac{x-x_0+|x-x_0|}{2}.$$
According to Trick 1 we know this can be uniformly approximated by polynomials.

3, We now focus on the bent at $c$ on the broken line.

See $\text{ReLU}(c)$, which has no effect on $[a,c]$ but changes the slope on $(c,b]$.

Thus we can figure out the exact expression of this broken-line, which only consists of polynomials and things like $|x-x_0|$.

---

---

Kernel Perturbation

By Landau, who shows us how to use kernel to fine-tune a function.

---

Essence:

1, Use a kernel to perturb it into a polynomial without changing any point’s being paramount at its own place.

---

Tricks:

1, We do a scaling and subtract a linear function to let $f(0)=f(1)=0\&f\equiv 0,x\notin [0,1]$.

2,

$Q_n(x):=c_n(1-t^2{)}^n$, which only makes sense at $0$, when $n$ is big enough.

And we want its integral $=1$, thus $c_n:=({\int }_{-1}^1(1-x^2{)}^n\text{d}x{)}^{-1}⩽\sqrt{n}$.

$⟹Q_n⩽\sqrt{n}(1-{\delta }^2{)}^n$.

3,

$P_n(x)={\int }_{-1}^{1}f(x+t)Q_n(t)\text{d}t,x\in [0,1].$

According to the property of $Q_n$, { P n } → f \{P_n\}\to f {Pn​}→f.

$P_n(x)={\int }_{-x}^{1-x}f(x+t)Q_n(t)\text{d}t={\int }_0^{1}f(m)Q_n(m-x)\text{d}m$.

Since $m$ will disappear after the integrating, this is an $x$-polynomial.

4,

$|P_n-f|⩽{\int }_{-1}^1|f(x+t)-f(x)|Q_n(t)\text{d}t$.

Then we use uniformly continuous.

Split the interval into ${\int }_{-1}^{-\delta }+{\int }_{\delta }^1+{\int }_{-\delta }^{\delta }$.

Then amplify to $\epsilon$.

---

---

Distribution is also a Kernel!

By Bernstein, who let us know that binomial distribution can be non-trivial.

---

Essence:

1, Binomial distribution has maximal expectancy at mid.

---

Tricks:

1, Do a scaling; let $f\in C[0,1]$.

2,

Construct kernel with binomial distribution: $P_{n,i}(x):=C_n^i{x}^i(1-x{)}^{n-i}$. Naturally, ${\sum }_{i=0}^n{P}_{n,i}=1$.

3,

$B_n(x):={\sum }_{i=0}^n{P}_{n,i}(x)f(\frac{i}{n}).$

Actually this expression has a form like expectancy.

4,

$|B_n(x)-f(x)|=\sum P_n(x)(f(\frac{k}{n})-f(x))=\cdots$.

Proof is like that of Landau.

---

Properties:

1, It’s exactly same at endpoints.

2, This has an explicit expression.

3, It converges very slowly.

4, Not a good approximation and some times funny.

Consider $x^2$ which is a polynomial per se but this method will not find it out.

Actually, consider such a set E n ( f ) : = { P ( x ) ∣ deg ⁡ P ⩽ n , ∣ f − P ∣ < ε , ∀ x } E_n(f):=\{P(x)|\deg P\leqslant n,|f-P|<\varepsilon,\forall x\} En​(f):={P(x)∣degP⩽n,∣f−P∣<ε,∀x}.

We want a minimal $\epsilon$, which polynomial is the best？

De facto, Tschebyscheff’s method is the best. It’s not unwarranted that we have a noun Tschebyscheff’s best approximation in high school.