1

Suppose $-\infty < x_{1}< x_{2}< \cdots < x_{n}< +\infty (n\geqslant 2)$ . And suppose a algebraic polynomial $C_{k}(x)$ $(k=1,2,\cdots ,n)$ ($degree\leqslant n-1$ ) satisfy :
$$C_{k}(x_{i})=\left\{\begin{matrix} 0, & i\neq k,\\ 1, & i=k, \end{matrix}\right. (i=1,2,\cdots ,n)$$

Prove : $C_{k}(x)+C_{k+1}(x)\geqslant 1$ , $x_{k}\leqslant x\leqslant x_{k+1}$ $(1\leqslant k\leqslant n-1)$ .

Let $C_{k}(x_{i})=a_{k}(x-x_{1})(x-x_{2})\cdots (x-x_{k-1})(x-x_{k+1})\cdots (x-x_{n})$ , $(k=1,2,\cdots ,n)$ .
$a_{k}=\frac{1}{(x-x_{1})(x-x_{2})\cdots (x-x_{k-1})(x-x_{k+1})\cdots (x-x_{n})} $ .
It satisfies the condition . But I can't prove the inequality .
Hope someone helps me .

Appriciate!

Luweizhao
  • 105
  • 1
    What have you tried? – DanLewis3264 Jul 22 '19 at 18:41
  • @LLWZ I know how to prove this, but I'm waiting to see what you've tried or done first to determine what issue(s) my answer would need to address. Please edit your question text to provide this, as well as possibly other context (e.g., where the problem comes from). Thanks. – John Omielan Jul 22 '19 at 18:52
  • Sorry for my poor english , I haven't solve it . – Luweizhao Jul 22 '19 at 19:00
  • @JohnOmielan The problem has an answer in the book , but it has an obvious mistake (He says the sum of function are monotonous on $[x_{k},x_{k+1}]$ ) . That doesn't convince me, so I'm looking for answers. – Luweizhao Jul 22 '19 at 19:14
  • @LLWZ Thanks for providing the extra info. I'm sorry, but I realized I didn't have a correct proof when I checked what I did before. However, I agree with you the answer at the book, i.e., the sum of the function is monotonous on $[x_k,x_{k+1}]$, is an obvious mistake. Also, I believe I've now given a correct proof in my answer. Two small points re: what you added are it should be $C_k(x)$, not $C_k(x_i)$; and you need to replace $x$ with $x_k$ in your $a_k$ definition. – John Omielan Jul 22 '19 at 21:38

2 Answers2

1

As the answer by Dunham has stated, the solution involves several aspects. First, let

$$P_k(x) = C_k(x) + C_{k+1}(x), \text{ for } 1\leqslant k\leqslant n-1 \tag{1}\label{eq1}$$

As you've indicated, you get

$$C_j(x) = \frac{\prod_{i=1,i\neq j}^{n}(x - x_i)}{\prod_{i=1,i\neq j}^{n}(x_j - x_i)}, \text{ for } 1 \le j \le n \tag{2}\label{eq2}$$

This $n-1$ degree polynomial is the only one of degree $\le n - 1$ which satisfies the required conditions (e.g., as discussed in the answer by Dan in Find N degree polynomial from N+1 points, and with more details in Polynomial interpolation).

As both $C_k(x)$ and $C_{k+1}(x)$ are $n-1$ degree polynomials, their sum of $P_k(x)$ is an up to $n-1$ degree polynomial. Also, $P_k(x_i) = 0$ for all $1 \le i \le n$ except for $i = k$ and $i = k+1$, i.e., for $n-2$ points. Also, note that $P_k(x_k) = P_k(x_{k+1}) = 1$.

Since $P'_k(x)$ is an up to $n-2$ degree polynomial, unless it's the $0$ polynomial, it can have at most $n-2$ roots. However, by the Mean value theorem, $P'_k(x)$ must have a root in each of $(x_i,x_{i+1})$ for all $1 \le i \le n - 1$ except for $i = k-1$ and $i=k+1$. This gives a total of $n-2$ points if $k = 1$ or $k = n - 1$, else $n-3$ points for all other $k$.

Consider that $P_k(x_m) \lt 1$ for some $x_k \lt x_m \lt x_{k+1}$. There are $4$ cases to consider, of which all but the first use the Intermediate value theorem:

  1. For $n = 2$, you get that $P_1(x)$ is a linear function. Since $P_1(x_1) = P_1(x_2) = 1$, this means it must be a constant function of $P_1(x) = 1$, so it's not possible for $P_1(x) \lt 1$.

  2. For $n \gt 2$ and $k = 1$, since $P_1(x_1) \gt P_1(x_m)$, $P_1(x_2) \gt P_1(x_m)$ and $P_1(x_2) \gt P_1(x_3)$, there are points $x_a$, $x_b$ and $x_c$ where $x_1 \lt x_a \lt x_m$, $x_m \lt x_b \lt x_2$ and $x_2 \lt x_c \lt x_3$ such that $P_1(x_a) = P_1(x_b) = P_1(x_c)$, so by the Mean value theorem, $P'_1(x)$ must at least $2$ roots in $(x_1,x_3)$. Since it has $n-3$ roots in $(x_3,x_n)$, this means $P'_k(x)$ must have at least $2 + (n-3) = n-1$ roots. However, since $P'_k(x)$ can have at most $n-2$ roots, this is not possible.

  3. For $n \gt 2$ and $1 \lt k \lt n - 2$, you can repeat the argument used in point $2$ above for both sides of $x_m$ to determine there must be at least $3$ roots of $P'_k(x)$ in $(x_{k-1},x_{k+2})$, but as there are $n-4$ other roots, the total of $3 + (n-4) = n - 1$ is once again too large.

  4. For $n \gt 2$ and $k = n-1$, you can apply the arguments used in point $2$ above for the left side of $x_m$ to, once again, determine that $P'_k(x)$ must have at least $n-1$ roots, which isn't possible.

Since all of the cases for $P_k(x_m) \lt 1$ for some $x_k \lt x_m \lt x_{k+1}$ have been shown to not be possible, this means that $P_k(x) \ge 1$ for all $x_k \le x \le x_{k+1}$ and $1 \le k \le n - 1$ as requested.

John Omielan
  • 47,976
  • Perfect proof ! There are some minor errors . In the paragraph "Since P′k(x) is" , the subscript of the open interval is incorrect . And at the third point , the interval I guess it should be $(x_{k-1},x_{k+2})$ . – Luweizhao Jul 23 '19 at 04:36
  • @LLWZ Thanks for the compliment & for pointing out my errors. I'm sorry about that as I did do a proof read, but I missed those issues. However, I've now fixed those errors. If you see any other mistakes, please let me know. Thanks. – John Omielan Jul 23 '19 at 04:43
0

Hint, rather than a complete answer. Try to prove by contradiction. Sketch some examples with 4-5 points to see where the problem is.

1) $P = C_k + C_{k+1}$ is polynomial of degree $\leq n-1$

2) $P$ has $n-2$ zeros at known locations

3) How many zeros does $P^{\prime}$ have?

Dunham
  • 3,297