Let $z_1,z_2,\dots,z_n\in\Bbb{C}$ be distinct and $w_1,w_2,\dots,w_n\in\Bbb{C}$ be arbitrary. Suppose $f, g$ are two polynomials of degree less than $n$ such that $$f(z_j)=w_j,\qquad g(z_j)=\bar{w}_j \qquad\text{for $1\leq j\leq n$}.$$ Define $\Omega(z)=\prod_{j=1}^n(z-z_j)$. The following puzzles me.

Question 1. Is this true? $$\sum_{k=1}^n\frac{\vert f^{\prime}(z_k)\vert^2}{\vert \Omega^{\prime}(z_k)\vert^2}\leq \sum_{k=1}^n\frac{\vert g^{\prime}(z_k)\vert^2}{\vert\Omega^{\prime}(z_k)\vert^2}.$$

EDIT. I'm sorry, one of the conditions was missing. To give credit, I'll leave the question as it stands and ask the correct one below.

Question 2. What if we insist that $f$ and $g$ have equal degrees?

  • $\begingroup$ Any further background where's this coming from? thanks. $\endgroup$
    – Suvrit
    Feb 5 '17 at 4:28
  • $\begingroup$ It was a technical need that occurred in some Monge-Ampere equations. $\endgroup$ Feb 5 '17 at 4:53
  • 2
    $\begingroup$ I am puzzled. Polynomials $\ f\ g\ $ appear in a symmetric way. Then why inequality and not equality? Indeed, if the theorem held for all $\ f\ g\ $ then there would be equality. $\endgroup$ Feb 5 '17 at 5:45
  • 1
    $\begingroup$ Well $f,g$ are uniquely defined by interpolation given the $z_j$'s and $w_j$'s, so $f=\bar{g}$ necessarily... $\endgroup$ Feb 5 '17 at 7:55
  • 2
    $\begingroup$ @PhilippeNadeau: it's not as easy as that; the $w_i$ are replaced by their complex conjugates but not the $z_i$, so the change in coefficients is a bit more mysterious. $\endgroup$ Feb 5 '17 at 9:41

For $n\ge 3$, let's take $z_j$, for $1\le j\le n$, be the $n$-th roots of unity, and $w_j:=z_j+4\bar{z_j}$, so that the assumption are satisfied by $$f(z):=z+4z^{n-1}$$ $$g(z):=4z+z^{n-1}.$$ However, $$ |g'(z_k)|=|4+(n-1) z_k^{-2}| \le n+3<\phantom{Z} $$ $$ \phantom{ZZZZ} <4n-5\le|1+4(n-1) z_k^{-2}|= |f'(z_k)| $$ for any index $k$ in the sums.

Rmk. For $n=2$, $f$ and $g$ are affine functions, with $|f'|=|g'|={|w_1-w_2|\over |z_1-z_2|}$, so that the claim is true.

  • 1
    $\begingroup$ Modified to fit question 2 (the customer is always right) $\endgroup$ Feb 5 '17 at 13:49
  • 2
    $\begingroup$ That was funny, Pietro! $\endgroup$ Feb 5 '17 at 14:22

This Matlab program shows that your conjecture is in general not true for $n\geq 3$

n = 3;

z = rand(1, n) + 1i * rand(1, n)
w = rand(1, n) + 1i * rand(1, n)

f = polyfit(z, w, n-1);
g = polyfit(z, conj(w), n - 1);

% note that Omega(z_i) = 0 and Omega(z) = z^n+q(z) where deg(q)<=n-1 and
% hence q(z_i)=-z_i^n
Omega = [1 -polyfit(z, z.^n, n - 1)];

df = polyder(f);
dg = polyder(g);
dOmega = polyder(Omega);

adf = abs(polyval(df, z));
adg = abs(polyval(dg, z));
adO = abs(polyval(dOmega, z));

  • $\begingroup$ I don't use mathlab. Do you mind giving the specific complex numbers that generate the example? $\endgroup$ Feb 5 '17 at 12:49
  • $\begingroup$ Pasting this into the Matlab console a couple of times gives a positive final value (11.1599) for z = [0.8143 + 0.3500i, 0.2435 + 0.1966i, 0.9293 + 0.2511i], w = [0.6160 + 0.8308i, 0.4733 + 0.5853i, 0.3517 + 0.5497i] $\endgroup$
    – J.J. Green
    Feb 5 '17 at 23:13

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.