The Stone-Weierstrass theorem states that for any continuous function $f$ on $[a,b]$ there exists a polynomial function $p$ such that $\|f-p\|<\varepsilon$ for any $\varepsilon>0$. Where $\|\cdot\|$ is the sup norm.

If I replace "a polynomial function" with "a polynomial function with rational coefficient" from the above statement, will the statement still hold true?


For sure it is true. For two polynomials $P_n,P_n^\prime$of the same degree $n$ with coefficients $p_0, \dots p_n$ and $p_0^\prime, \dots , p_n^\prime$ respectively, you have $$\Vert P_n -P_n^\prime\Vert_\infty =\sup\limits_{t \in [a,b]} \vert P_n(t)-P_n^\prime(t)\vert \le c \left(\sup\limits_{1\le i\le n}\vert p_i -p_i^\prime\vert\right)$$

with $c=\sup\limits_{t \in [a,b]} \sum_{k=0}^n \vert t \vert^k$.

Hence if you have a real polynomial $P_n$ at a distance less than $\epsilon/2$ of $f$, you can find a polynomial of same degree with rational coefficients $P_n^\prime$ at a distance less then $\epsilon/2$ of $P_n$. And you can conclude with triangular inequality.





$p(x) \in \Bbb R[x], \tag 1$

then given any $\epsilon$ there exists a polynomial

$q(x) \in \Bbb Q[x] \tag 2$

such that

$\vert p(x) - q(x) \vert < \epsilon, \; x \in [a, b]; \tag 3$

for, writing

$p(x) = \displaystyle \sum_0^n p_i x^i, \; p_i \in \Bbb R, \tag 4$


$q(x) = \displaystyle \sum_0^n q_i x^i, \; q_i \in \Bbb R, \tag 5$

then, assuming $a \ne b$,

$\vert p(x) - q(x) \vert = \vert \displaystyle \sum_0^n p_i x^i - \sum_0^n q_i x^i \vert \le \sum_0^n \vert p_i - q_i \vert \vert x \vert^i \le \sum_0^n \vert p_i - q_i \vert (\max(\vert a \vert, \vert b \vert))^i; \tag 6$

given $p_i$, $0 \le i \le n$, we may choose $q_i$ such that

$\vert p_i - q_i \vert < \dfrac{\epsilon}{(n + 1)(\max_{0 \le i \le n}(\max(\vert a \vert, \vert b \vert))^i)}; \tag 7$


$\vert p(x) - q(x) \vert \le \displaystyle \sum_0^n \vert p_i - q_i \vert (\max(\vert a \vert, \vert b \vert))^i$ $< \displaystyle \sum_0^n \dfrac{\epsilon(\max(\vert a \vert, \vert b \vert))^i)}{(n + 1)(\max_{0 \le i \le n}(\max(\vert a \vert, \vert b \vert))^i)}$ $\le \dfrac{\epsilon(\max(\vert a \vert, \vert b \vert))^i)}{(\max_{0 \le i \le n}(\max(\vert a \vert, \vert b \vert))^i)} \le \epsilon. \tag 8$

It follows that for every $\epsilon > 0$ and any $p(x)$ as in (1), there exists $q(x)$ as in (2) with $\vert p(x) - q(x) \vert < \epsilon$ on $[a, b]$. Now replacing $\epsilon$ with $\epsilon / 2$ and choosing $p(x) \in \Bbb R[x]$ such that

$\vert f(x) - p(x) \vert < \dfrac{\epsilon}{2}, \; x \in [a, b], \tag 9$

we may find some $q(x) \in \Bbb Q[x]$ such that

$\vert p(x) - q(x) \vert < \dfrac{\epsilon}{2}, \tag{10}$

and combining (9) and (10) we have

$\vert f(x) - q(x) \vert = \vert f(x) - p(x) + p(x) - q(x) \vert$ $\le \vert f(x) - p(x) \vert + \vert p(x) - q(x) \vert < 2 \dfrac{\epsilon}{2} = \epsilon, \; x \in [a, b], \tag{11}$


$\Vert f(x) - q(x) \Vert < \epsilon. \tag{12}$


Yes this is true. We know that we can approximate $f$ arbitrarily closely by a polynomial $p = \sum_k a_k x^k$ with real coefficients. Let $q = \sum_k c_kx^k$ be a polynomial with rational coefficients. Then for any $x \in [a,b]$ $$ |p(x)-q(x)| \leq \sum_{k=1}^n |a_k-c_k||x^k| \leq \sum_{k=1}^n |a_k-c_k| b^k $$ By density of rationals, we can make $|a_k-c_k| < \frac{\epsilon}{nb^k}$.


Yes. We know there is some $p$ of degree $n$ with $||f-p|| < \epsilon$. You can show that the map $f: \mathbb{R}^{n+1} \rightarrow \mathbb{R}$ that sends $(a_0, \ldots, a_n)$ to $||f - p||$ where $p(x) = \sum_{i=0}^n a_i x^i$ is continuous, so $f^{-1}( (0, \epsilon))$ is a nonempty open set; since rationals are dense in $\mathbb{R}^{n+1}$, there must be a rational point in $f^{-1}((0, \epsilon))$.


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.