Given a function with nice properties, e.g. $f \in \mathcal{C}^k([a,b])$, are there any guarantees, i.e. upper bounds, on the $L^2$ error

\begin{equation} min_{g \in \Pi_n} \|f-g\|_{L^2} \end{equation}

for the optimal approximation with a polynomial of fixed degree $n$? If there are no such guarantees, is there a result that proves the existence of functions with arbitrarily bad approximations?

I am aware of the theorem by Bernstein which establishes the existence of arbitrarily bad functions to approximate with regards to the $L^{\infty}$ norm.


We can construct one explicitly. We use the interval $[-1,1]$ to be specific. Let $P_i$ be the $i^{th}$ degree Legendre polynomial. Because they are orthonormal with weight $1$ the best $n$th degree fit to $f$ is $\sum_{i=0}^n \left(\int_{-1}^1 f(x)P_i(x)dx\right)P_i(x)$ Now let $f(x)=kP_{n+1}(x)$. As the polynomials are orthornormal, the best $n$th degree approximation to $f$ is the zero polynomial. As $f$ has non-zero norm, we can choose $k$ large enough to make the error of the approximation as large as we want.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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