0
$\begingroup$

Given two polynomials $p$ and $q$ over reals and being guaranteed that both have all roots real I want to know if there is any characterization of the solutions of the equation $\frac{p'}{p} = \frac{q'}{q}$.

  • In particular I want to know if there is a known statement about when can the above equation have a root which is $ > max\{maxroot(p),maxroot(q)\}$.

  • And if a root of the above kind exists then can one say something about the value of $\frac{p'}{p}$ ( $= \frac{q'}{q}$) at that point?

  • Also if a root of this equation exists which is $> max\{maxroot(p),maxroot(q)\}$ then can one somehow estimate where this root is by say knowing $p$ and $q$.


  • If necessary feel free to assume that the roots of $p$ and $q$ are known exactly.

  • If necessary assume that $p$ and $q$ are both monic polynomials of the same degree.

$\endgroup$

1 Answer 1

2
$\begingroup$

If $p(z) = a \prod_{j=1}^d (z - r_j)$ and $q(z) = b \prod_{j=1}^e (z - s_j)$, $r_1 \le r_2 \le \ldots \le r_d$, $s_1 \le s_2 \le \ldots \le s_e$, then let $$ f(z) = \dfrac{p'}{p} - \dfrac{q'}{q} = \sum_{j=1}^d \dfrac{1}{z-r_j} - \sum_{j=1}^e \dfrac{1}{z-s_j} = \sum_{k=0}^\infty \dfrac{m_k}{z^{k+1}}$$ for $z > R = \max( r_d, s_e)$, where $m_k = \sum_{j=1}^d r_j^k - \sum_{j=1}^e s_j^k$. We may assume wlog the sets $\{r_j\}$ and $\{s_j\}$ are disjoint.

The sign of $f(z)$ as $z \to R+$ is positive if $r_d > s_e$, negative if $r_d < s_e$. The sign as $z \to +\infty$ is that of the first nonzero coefficient $m_k$ (which must exist unless $d=e=0$). A sufficient condition for the existence of a zero of $f$ in $(R,\infty)$ is that these signs are different.

Moreover, by bounding the tail of the series we can place bounds on the location of any zeros. Thus suppose the first nonzero coefficient is $m_k$. We have $|m_j| \le (d+e) M^j$ where $M = \max(|r_1|,\ldots,|r_d|,|s_1|,\ldots,|s_e|)$, so that for $z > M$, $$|f(z)| \ge \dfrac{|m_k|}{z^{k+1}} - \sum_{j=k+1}^\infty (d+e) \dfrac{M^j}{|z|^{j+1}} = \dfrac{|m_k|}{z^{k+1}} - (d+e) \dfrac{M^{k+1}}{|z|^{k+2}(1 - M/|z|)} $$

so any zero must satisfy $$ z \le M + \dfrac{(d+e) M^{k+1}}{|m_{k}|}$$

EDIT: $m_0 = d - e$, so that's the first nonzero coefficient if $d \ne e$. If $d = e$, then at least one of $m_1, \ldots, m_d$ is nonzero. To see this, consider the power-sums $M_i(X_1, \ldots, X_d) = \sum_{j=1}^d X_j^i$ for $ i=1\ldots d$ as symmetric polynomials in $X_1, \ldots, X_d$. It is well-known that these form an algebraic basis for the symmetric polynomials in $X_1, \ldots, X_d$. Let $Q(X_1,\ldots, X_d,y) = \prod_{j=1}^d (X_j - y)$. This is nonzero if and only if $y \notin \{X_1, \ldots, X_d\}$, and its $k$'th derivative with respect to $y$ is nonzero if and only if at most $k$ of $X_1, \ldots, X_d$ are equal to $y$. $Q(\cdot, \ldots, \cdot, y)$ and its derivatives with respect to $y$ are symmetric polynomials in $X_1, \ldots, X_d$, so they are determined by the power-sums $M_i(X_1, \ldots, X_d)$. Thus the only way to have $m_j = M_j(r_1, \ldots, r_d) - M_j(s_1, \ldots, s_d) = 0$ for $j=1\ldots d$ is that $(r_1, \ldots, r_d)$ is a permutation of $(s_1, \ldots, s_d)$.

$\endgroup$
8
  • $\begingroup$ Thanks for the insights! (1) Finding this smallest $k$ for which $m_k \neq 0$ seems like an indefinite process right? Can this be done in a finite number of steps? (2) For my immediate purposes I can put in some further restrictions like $a=b=1$, $d=e$ and that the intervals $[ min(r_i,s_i),max(r_i,s_i)]$ are disjoint for all $i$ unless some repeated root of $p$ is common with $q$. Do these help refine anything? $\endgroup$
    – user6818
    Commented May 29, 2015 at 7:17
  • $\begingroup$ And in the last inequality why are you thinking in terms of the modulus of $f(z)$? One can do the argument will the absolute values removed? (but the hard part of the argument is about how does one estimate the value $k$?) $\endgroup$
    – user6818
    Commented May 29, 2015 at 9:13
  • $\begingroup$ Also can you explain why is $sgn(f(max(r_d,s_e)^+)) = sgn(r_d-s_e)$ ? $\endgroup$
    – user6818
    Commented May 29, 2015 at 9:40
  • $\begingroup$ If $d \ne e$, $m_0 \ne 0$. Otherwise, I suspect the smallest $k$ can't be greater than $d$. $\endgroup$ Commented May 29, 2015 at 15:31
  • $\begingroup$ Why should $d$ be special? You mean if $d=e$ then $k < d=e$ ? Also can you explain why you claimed that $sgn(f([max(r_d,s_e) ]^+ ) ) = sgn(r_d - s_e)$ $\endgroup$
    – Student
    Commented May 29, 2015 at 16:14

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .