2

Is there is a positive integer $m$, depending only on $n$, such that for any strictly increasing integer sequence $a _{1}, a _{2}, \dots, a _{n}$, there is some polynomial $f(x)$ of degree at most $m$ with rational coefficients, such that $$f(i) = a_{i} \quad (i = 1, 2, \dots, n)$$ and $f(x)$ is monotonic increasing on $[ 1, n]$?

It feels like using Lagrangian interpolation, $$f(x)=\sum_{i=1}^{n}\left(f(i)\prod_{j\neq i}\dfrac{(x-j)}{(i-j)}\right)=\sum_{i=1}^{n}\left(a_{i}\prod_{j\neq i}\dfrac{(x-j)}{(i-j)}\right)$$ but I find there's no guarantee that this polynomial $f(x)$ will be monotone increasing on $[1,n]$.

user125932
  • 8,813
math110
  • 93,304
  • The Langrangian interpolation is highly unlikely to monotone. Far more likely, it will swing back and forth between the points. This is its most common behavior and the sequence being increasing by itself is not enough to prevent it. Lagrange is the minimal degree interpolating polynomial. To pull this off, you are going to need a much higher degree polynomial. In general, high enough that you can fit an inflection point between each pair of adjacent supports. – Paul Sinclair Oct 14 '19 at 23:18

1 Answers1

7

No, for $n \geq 3$ there is no such $m$ which applies to all sequences $a_1, \dots, a_n$.

Proposition. If $f$ is a polynomial of degree $d$ with real coefficients, such that $f$ is monotone increasing on $[0, 2]$ with $f(0) = 0$ and $f(1) = 1$, then we must have $f(2) \leq (2d)^{d+1}$.

Proof: Consider the Lagrange interpolation formula for $f(x)$, where we interpolate at the $d+1$ points $0, \frac{1}{d}, \dots, \frac{d-1}{d}, 1$: $$f(x) = \sum_{i=0}^d \left(f(i/d) \prod_{0 \leq j \leq d, j \neq i} \frac{x - j/d}{i/d - j/d}\right)$$

For all $i, j$ we have $0 \leq f(i/d) \leq 1$, $|2 - j/d| \leq 2$, and $|i/d - j/d| \geq 1/d$ (when $i \neq j$), so it follows that $$|f(2)| \leq \sum_{i=0}^d \left(|f(i/d)| \prod_{0 \leq j \leq d, j \neq i} \frac{|2 - j/d|}{|i/d - j/d|}\right) \leq (d+1) \frac{2^d}{(1/d)^d} \leq (2d)^{d+1}$$ as desired.

Then when $n = 3$, for any fixed $m$, using this lemma we see that any polynomial with rational coefficients which satisfies $f(1) = 0$, $f(2) = 1$, and $f(3) = (2m)^{m+1}+1$ and is monotone on $[1, 3]$ must have degree at least $m+1$. This means there is no $m$ for which we can always find a polynomial of degree $\leq m$ to fit any given sequence $a_1, a_2, a_3$.

user125932
  • 8,813
  • at last,I can't understand why "This means there is no mm for which we can always find a polynomial of degree $\le m$,to fit any given sequence $a_{1},a_{2},a_{3}?$ can you explain? – math110 Oct 18 '19 at 15:00
  • I've edited to clarify the wording -- does that help? – user125932 Oct 18 '19 at 17:01