Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Assume one wants to minimize the distance between $f(x)=\ln x$ and $P_n(t)$ where $P_n$ denotes a polynomial of degree $n$. Etc $P_1 = ax 0+ b$. One way to judge whether the polynomial is a good approximation is to minimize the integral

$$ E_n = \int_0^1 \left| \ln(t) -P_n(t)\right|^2\mathrm{d}t $$

The polynomials I have found so far are

$$\begin{align} P_1(t) & = 3x - \frac{5}{2} \\ P_2(t) & = -5x^2 + 8x - \frac{10}{3} \\ P_3(t) & = \frac{35}{3}x^3 - \frac{45}{2}x^2 + 15x - \frac{47}{12} \end{align}$$

Now after some numerical calculations, it seems that $E_n =1/(n+1)^2$. But why?

share|improve this question
    
I think you can consider the orthogonal projection of ln(x) onto the subspace generated by polynomial of degree lesser or equal to n. –  Ben Nov 22 '12 at 14:56
    
If you take $$P_N(x)=-1+\sum_{m=1}^{N}\frac{(-1)^{m+1}(2m+1)}{m(m+1)!}\,\frac{d^m}{dx^m}(x(1‌​-x))^m$$ you have $$\|\log(x)-P_N(x)\|_2^2=\frac{1}{(N+1)^2},$$ proof below. –  Jack D'Aurizio Nov 22 '12 at 18:50
add comment

1 Answer 1

up vote 5 down vote accepted

Let $L_n(x)$ be the $n$-th Legendre polynomial. These polynomials are a complete orthogonal base of $\mathcal{L}^2([-1,1])$ with respect to the inner product $$<f,g>=\int_{-1}^{1}f(x)\,g(x)\,dx$$ and they satisfy $$\int_{-1}^{1}L_n(x)\,L_m(x)\,dx=\frac{2}{2n+1}\,\delta_{m,n}.$$ Since, by the Rodrigues formula, $$ L_n(x) = \frac{1}{2^n\,n!}\cdot\frac{d^n}{dx^n}(1-x^2)^n,$$ it is not difficult to compute $$ \int_{-1}^{1} L_n(x)\,\log(1+x)\,dx = (-1)^{n+1}\frac{2}{n(n+1)}, $$ from which it follows that $$ \log(1+x) = -\frac{\log 2}{2}+\sum_{n=1}^{+\infty}\frac{2n+1}{n(n+1)}L_n(x).$$ Now, if you take $$ P_N(x) = -\frac{\log 2}{2}+\sum_{n=1}^{N}\frac{2n+1}{n(n+1)}L_n(x).$$ as a polynomial approximation of $\log(1+x)$ with degree $N$, by the orthogonality of the Legendre polynomials you get: $$ \|P_N(x)-\log(1+x)\|_2^2 = \sum_{n>N}\frac{(2n+1)^2}{n^2(n+1)^2}\|L_n(x)\|_2^2=$$ $$ = 2\sum_{n>N}\frac{2n+1}{n^2(n+1)^2}=2\sum_{n>N}\left(\frac{1}{n^2}-\frac{1}{(n+1)^2}\right)=\frac{2}{(N+1)^2},$$ with the same magnitude of the error of the best polynomial approximation of degree $N$.

In a similar manner, the polynomials $$J_n(x)=\frac{1}{n!}\frac{d^n}{dx^n}(x(1-x))^n$$ form a complete orthogonal base of $\mathcal{L}^2([0,1])$ with the standard inner product, $\|J_n\|_2^2 = \frac{1}{2n+1}$ and $$\int_{0}^{1}\log(x)\,J_n(x)\,dx = \frac{(-1)^{n+1}}{n(n+1)},$$ so if you take $$P_N(x)=-1+\sum_{m=1}^{N}\frac{(-1)^{m+1}(2m+1)}{m(m+1)}\,J_m(x)$$ you have $$\|\log(x)-P_N(x)\|_2^2=\frac{1}{(N+1)^2}.$$

share|improve this answer
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.