MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

It is known from finite dimensionality of $P_r$, the space of all polynomials of degree less than or equal to $r$, that $P_r$ is complete with respect to uniform norm.

Considering $R_{m,n}[a,b]=\{p/q: p \in P_m[a,b],q\in P_n[a,b], q|_{[a,b]}\neq 0\}$ (which is not a linear space) endowed with the uniform metric, is it a complete metric space? if not, exhibit a Cauchy sequence that is not convergent.

share|cite|improve this question
This question does not seem too trivial to me. – Joseph Van Name Jun 13 '13 at 12:08
I gave it some thought, but I did not yet come up with an easy way to do it. Rational functions are very special things: meromorphic on the Riemann sphere. With at most a fixed number of poles (counted with multiplicity). – Gerald Edgar Jun 14 '13 at 14:30
Maybe not trivial, but the imperative formulation still suggests it's an exercise and thus off topic here. – Noam D. Elkies Jun 14 '13 at 21:42
I have my doubts it's homework. The imperative formulation might be a result of less than perfect command of English -- we've talked about this sort if thing before at MO. – Todd Trimble Jun 15 '13 at 14:28

At the risk that I'm doing somebody's homework...

Yes, $R_{m,n}[a,b]$ is complete with respect to the uniform metric $\left\| \cdot \right\|$.

Let $f \in {\cal C}([a,b])$ be a uniform limit of rational functions $f_j = p_j \phantom./\phantom. q_j \in R_{m,n}$ with each $p_j$ (respectively $q_j$) a polynomial of degree at most $m$ (resp. $n$), and $q_j(x) \neq 0$ for all $x \in [a,b]$. We shall show that $f \in R_{m,n}[a,b]$.

We would like to write $f = p/q$ with $p = \lim_j p_j$ and $q = \lim_j q_j$, but this doesn't work in general because $f_j$ does not determine $p_j,q_j$ uniquely: we can multiply $p_j,q_j$ by the same scalar, and if $f_j$ is actually a quotient of polynomials of degrees strictly less than $m$ and $n$ then we could even introduce a nonconstant common factor into $p_j$ and $q_j$.

We deal with the first ambiguity by scaling each $p_j$ and $q_j$ so that $\left\| q_j \right\| = 1$. Then the $q_j$ are all on the unit sphere in the finite-dimensional space $P_n[a,b]$, which is compact. This lets us finesse the second ambiguity by passing to a subsequence that converges in ${\cal C}([a,b])$.

Let $q$, then, be the limit of such a subsequence $\lbrace q_k \rbrace$, and let $p = f q$. Then $p$ is the uniform limit of polynomials $p_k = f_k q_k$, because $f$ and $q$ are uniform limits of $f_k$ and $q_k$ respectively. But each $p_k$ is in the finite-dimensional vector space $P_m$, which is complete in ${\cal C}([a,b])$; so the uniform limit $p$ is also a polynomial, of degree at most $m$.

Let $f_\infty \in R_{m,n}$ be the rational function $p/q$, reduced to lowest terms. We claim that $f = f_\infty$ on $[a,b]$. This is not quite immediate from what we've done so far, because $q$ might have a zero $x_0 \in [a,b]$ even though none of the $q_k$ does. [For example, let $m=n=2$ and $f_k = 1 + 1/(k^2 (x-x_0)^2 + k)$.] Still, since $f_k = p_k/q_k$ with $p_k \rightarrow p$ and $q_k \rightarrow q$, it does follow that $f_\infty(x) = f(x)$ for all $x \in [a,b]$ where $q(x) \neq 0$. Since $f$ is continuous on $[a,b]$, it follows that $f_\infty(x) = f(x)$ even at each of the finitely many $x$ where $q(x) = 0$. QED

share|cite|improve this answer

Not my area of expertise, but appears to be well-known in the literature that this is a complete metric space. From here for instance, we have that any continuous function $f$ on $[-1, 1]$ has a best uniform approximation by rational functions $r = p/q$ with $\deg(p) \leq m$, $\deg(q) \leq n$, and $q$ non-vanishing on $[-1, 1]$, in the sense that for some such (unique) $r^*$ we have

$$\|f - r^*\| \leq \|f - r\|$$

in sup norm for any other such $r$. Hence the infimum of $\|f - r\|$ cannot be zero, unless of course $f = r^*$ already belongs to the space.

share|cite|improve this answer

Your Answer


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.