Let us consider monic polynomials of $n^{\text{th}}$ degree $p(t) = t^n + a_{n-1} t^{n-1} + \dots + a_1 t + a_0$ with real coefficients. We know the roots are continuous functions of the coefficients. Let us define a map $F: \mathbb R^n \to \mathbb R_+^n$ by \begin{align*} F: (a_{n-1}, \dots, a_0) \mapsto (r_1, \dots, r_n) \mapsto (|r_1|, \dots, |r_n|), \end{align*} where $r_j = r_j(a_{n-1}, \dots, a_0)$ are roots of the polynomial and $|\cdot|$ denotes the modulus of a complex number. Clearly $F$ is a continuous function. Let us take the Euclidean norm on $\mathbb R^n$. Suppose for some fixed coefficients $b = (b_{n-1}, \dots, b_0)$, the roots are all nonzero and let $(c_1, \dots, c_n)$ be the roots. if we take an open ball $B_b(\varepsilon)$ around $b$, will the image $F[ B_b(\varepsilon) ]$ contain a cube, i.e., $(|c_1|-\varepsilon', |c_1|+\varepsilon') \times (|c_2|-\varepsilon', |c_2|+\varepsilon') \times \dots \times (|c_n|-\varepsilon', |c_n|+\varepsilon')$ for some possible smaller $\varepsilon'$?

  • $\begingroup$ Your map $F$ is not well-defined: the roots are only defined as an unordered collection, not as an ordered $n$-tuple. It would be well-defined (and continuous) if you changed the codomain to $\mathbb{R}^n_+/S_n$. $\endgroup$ – Eric Wofsey Jul 8 '18 at 6:57
  • $\begingroup$ You could still consider your $F$ as a "multi-valued function" and ask your question about the image $F[B_b(\epsilon)]$, though. $\endgroup$ – Eric Wofsey Jul 8 '18 at 6:58

As I commented, your function $F$ is not well-defined. However, your question still makes sense without it. You can ask, given a multiset of nonzero complex numbers $c_1,\dots,c_n$ invariant under conjugation and $\epsilon>0$, does there exist $\epsilon'>0$ such that for any $r_1,\dots,r_n\in\mathbb{R}$ with $|r_k-|c_k||<\epsilon'$ for each $k$, there exist $d_1,\dots,d_n\in\mathbb{C}$ also invariant under conjugation such that $|d_k|=r_k$ for each $i$ and coefficients of the monic polynomial of with the $d_k$ as roots are within $\epsilon$ of the coefficients of the monic polynomial with the $c_k$ as roots. (The invariance under conjugation condition here corresponds to the requirement that the polynomial has real coefficients.)

The answer to this question is no. For instance, let $n=2$, and $(c_1,c_2)=(i,-i)$ and consider $r_1=1+\epsilon'/2$ and $r_2=1$. Notice then that if $|d_k|=r_k$ for $k=1,2$, then $d_1$ and $d_2$ cannot be conjugates of each other, and so they must both be real in order to be invariant under conjugation. It follows that there are no such $d_1$ and $d_2$ that give rise to a polynomial close to the polynomial $t^2+1$ with roots $c_1$ and $c_2$.

Or, in more concrete terms, there is no polynomial with real coefficients close to $t^2+1$ whose roots have distinct absolute values, since any nearby polynomial has no real roots and hence the roots must be a complex conjugate pair.

  • $\begingroup$ Thanks. Will the ''conclusion'' hold: if we let the roots to be identical real numbers, i.e., $c=(\beta, \beta, \dots, \beta)$? Here by conclusion I mean: an open cube $(|\beta| - \varepsilon', |\beta| + \varepsilon' )^n$ such that all the conjugate invariant subset is contained in the image of $F[B_b(\varepsilon)]$? $\endgroup$ – user1101010 Jul 8 '18 at 16:04
  • $\begingroup$ Yes, it is true if the roots are all real. This follows immediately from the fact that the function taking a collection of roots to the coefficients of the corresponding polynomial is continuous (since it's just given by the elementary symmetric polynomials). So, just pick a collection of real roots near $\beta$ with the desired absolute values, and if they are sufficiently near $\beta$, the resulting polynomial will be sufficiently near the initial polynomial. $\endgroup$ – Eric Wofsey Jul 8 '18 at 16:30

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.