13

Given some roots : $r_1,r_2,\ldots,r_n$, how can we reconstruct polynomial coefficients?

I know the Horner scheme and that we can just go backwards receiving those coefficients.

But I'm curious if there's some other nice solution to this problem.

user26857
  • 53,190
  • 7
    I'm not sure I'm answering your question, but to get the coefficients from the roots, just expand $(X-r_1) (X-r_2) \ldots (X-r_n)$. You get $$(X-r_1) (X-r_2) \ldots (X-r_n) = X^n - (r_1 + r_2 + \cdots + r_n) X^{n-1}+ \cdots + (-1)^n r_1 r_2 \ldots r_n X^0$$ The expressions you find are called symmetric polynomials. – Joel Cohen Oct 03 '11 at 15:21
  • 3
  • 2
    It's not clear to me how you can take the Horner scheme "backwards" to get the coefficients. – Thomas Andrews Oct 03 '11 at 15:27
  • Joel could you elaborate hows this equation goes further? I can't get it – Krzysztof Lewko Oct 03 '11 at 15:52
  • 2
    @Chris : The $X^{n-2}$ coefficient is $\sum_{1 \le i, j \le n} r_i r_j$ (sum of products of $2$ roots), then the $X^{n-3}$ coefficient $-\sum_{1 \le i, j, k \le n} r_i r_j r_k$ (sum of products of $3$ roots), the $X^{n-4}$ coefficient $\sum_{1 \le i, j, k, l \le n} r_i r_j r_k r_k$ (sum of products of $4$ roots) and so on. The $X^{n-k}$ coefficient is $(-1)^k\sum_{1 \le i_1, i_2, \ldots, i_k \le n} r_{i_1} r_{i_2} \ldots r_{i_k}$ (sum of products of $k$ roots). – Joel Cohen Oct 03 '11 at 16:37
  • Please see http://en.wikipedia.org/wiki/Vandermonde_matrix – leo Oct 03 '11 at 18:52
  • Thanks Joel, now that's clear. – Krzysztof Lewko Oct 03 '11 at 20:52

2 Answers2

17

Knowing the location of roots really only pins down a family of polynomials, which vary by multiplicative constants. If you know the roots of $P(x)$ are $r_1, r_2, \dots, r_n $ then you can write the polynomial as $$ P(x) = a (x-r_1)(x-r_2) \cdots (x-r_n) $$

where $a$ is some non-zero constant. However, we can not be more precise than that. We can pin down $a$ if we know the value of $P$ are any other point. If we knew $a$, then simply expanding the right hand side would produce the polynomial in a form where you could read off the constants.

user26857
  • 53,190
Ragib Zaman
  • 36,065
  • 2
    As a tiny note: the requirement of an $a$ along with the $n$ roots is very much consistent with the fact that one requires $n+1$ conditions to uniquely determine a polynomial. – J. M. ain't a mathematician Oct 04 '11 at 00:42
9

The polynomial is $P(z) = a_0 (z-r_1)\times (z-r_2) \times \cdots \times (z-r_n)$. Now labeling coefficients as $P(z) = a_0 z^n + a_1 z^{n-1} + \ldots + a_n$ we have $$ a_1 = -a_0 ( r_1 + r_2 + \ldots + r_n) \qquad a_n = (-1)^n a_0 r_1 \times r_2 \times \cdots \times r_n $$ In general: $$ a_k = \frac{(-1)^k}{k!} a_0 \sum_\pi r_{\pi(1)} \times \cdots \times r_{\pi(k)} $$ where the sum is over permutations $ \pi \in \mathfrak{S}_n$.

Sasha
  • 71,686