Suppose I have a polynomial (of any order) and I'm not able to calculate the roots. Is there a way to get at least some information about the roots such as how many of them are complex, negative or positive? For example, I can safely identify the behavior (and therefore roots' character) of $f(x)=ax+b$ or even a quadratic expression just by inspection.
I'm aware of Descartes' sign rule http://en.wikipedia.org/wiki/Descartes%27_rule_of_signs but it apparently provides only an upper bound on the number of positive/negative roots. Is there something more general giving an exact number (of positive roots) preferably without using the methods of calculus?
If $P=\sum_{k=0}^n a_kX^k$, then the roots of $P$ are bounded by $\displaystyle \max\{1,\sum_{k=0}^{n-1}\left|\frac{a_k}{a_n}\right|\}$
Note that the proof adapts to complex polynomials as well.
Suppose $P=\sum_{k=0}^n a_kX^k$ has complex roots $\alpha_1,\ldots,\alpha_n$.
Let $i\in \{1,\ldots,n\}$
$\displaystyle P(\alpha_i)=0=\sum_{k=0}^n a_k\alpha_i^k$
Two cases come up:
$|\alpha_i|\leq 1$
$|\alpha_i|> 1$
In this last case, write $\displaystyle |\alpha_i|=|\sum_{k=0}^{n-1}\frac{a_k}{a_n}\alpha_i^{k-n+1}|\leq \sum_{k=0}^{n-1}\left|\frac{a_k}{a_n}\alpha_i^{k-n+1}\right|$
By assumption, $\forall k\in\{0,\ldots,n-1\},|\alpha_i^{k-n+1}|\leq 1$
Hence $\displaystyle |\alpha_i|\leq \sum_{k=0}^{n-1}\left|\frac{a_k}{a_n}\right|$
Therefore, if $P=\sum_{k=0}^n a_kX^k$, then the roots of $P$ are bounded by $\displaystyle \max\{1,\sum_{k=0}^{n-1}\left|\frac{a_k}{a_n}\right|\}$