Polynomial Approximations to Common Functions

On this page, we list many functions which can be uniformly approximated by low-degree polynomials on bounded domains. For all results stated here, we let f(x)f(x) denote the function we want to approximate, dd denote the degree of the polynomial approximation, and ε\varepsilon denote the uniform approximation error, so that

supx[a,b]f(x)p(x)εfor some deg(p)d \sup_{x\in[a,b]}\left|f(x) - p(x)\right| \leq \varepsilon \hspace{1cm} \text{for some deg}(p) \leq d

Theorem 1: Approximate Monomial
Theorem

Fix positive integers dd and qq, and suppose we want to approximate f(x)=xqf(x)=x^q for x[1,1]x\in[-1,1]. Then there exists a polynomial p(x)p(x) with degree dd such that

supx[1,1]p(x)xq2ed2q \sup_{x\in[-1,1]} \left|p(x)-x^q\right| \leq 2e^{-\frac{d^2}{q}}

Or, equivalently, there exists a degree d=2qln(2ε)d=\sqrt{2q\ln(\frac2\varepsilon)} polynomial p(x)p(x) such that

supx[1,1]p(x)xqε \sup_{x\in[-1,1]} \left|p(x)-x^q\right| \leq \varepsilon

(Theorem 3.3 from Sachdeva Vishnoi (2013))

Theorem 2: Approximate Negative Exponential
Theorem

Suppose we want to approximate f(x)=exf(x)=e^{-x} for x[0,b]x\in[0,b]. Then there exists a degree d=O(max{b,log(1ε)}log(1ε))d=O(\sqrt{\max\{b,\log(\frac1\varepsilon)\} \cdot \log(\frac1\varepsilon)}) polynomial pp such that

supx[0,b]p(x)exε \sup_{x\in[0,b]} \left|p(x)-e^{-x}\right| \leq \varepsilon

(Theorem 4.1 from Sachdeva Vishnoi (2013))

This result for negative exponential can be very easily generalized to other functions, given Theorem 1, so if you need to make your own polynomial approximation, consider reading Chapter 4 from Sachdeva Vishnoi (2013).

Bibliography